Accepted Manuscript

Artificial neural network and particle swarm optimization for removal of methyl
orange by Gold nanoparticles loaded on activated carbon and Tamarisk

M. Ghaedi, A.M. Ghaedi, A. Ansari, F. Mohammadi, A. Vafaei

PII: S1386-1425(14)00748-3
DOI: http://dx.doi.org/10.1016/j.saa.2014.04.175
Reference: SAA 12139

To appear in: Spectrochimica Acta Part A: Molecular and Biomo-

lecular Spectroscopy

Received Date: 3 March 2014

Revised Date: 20 April 2014
Accepted Date: 23 April 2014

Please cite this article as: M. Ghaedi, A.M. Ghaedi, A. Ansari, F. Mohammadi, A. Vafaei, Artificial neural network
and particle swarm optimization for removal of methyl orange by Gold nanoparticles loaded on activated carbon
and Tamarisk, Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy (2014), doi: http://
dx.doi.org/10.1016/j.saa.2014.04.175

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 Artificial neural network and particle swarm optimization for removal of methyl orange
by Gold nanoparticles loaded on activated carbon and Tamarisk

M. Ghaedia,*, A.M. Ghaedib,**, A. Ansarib, F. Mohammadib, A. Vafaeib

a
Chemistry Department, Yasouj University, Yasouj 75918-74831, Iran
b
Department of Chemistry, Gachsaran Branch, Islamic Azad University, P.O. Box 75818-
63876, Gachsaran, Iran,

Abstract

The influence of variables, namely initial dye concentration, adsorbent dosage (g), stirrer speed

(rpm) and contact time (min) on the removal of methyl orange (MO) by gold nanoparticles

loaded on activated carbon (Au-NP-AC) and Tamarisk were investigated using multiple linear

regression (MLR) and artificial neural network (ANN) and the variables were optimized by

partial swarm optimization (PSO). Comparison of the results achieved using proposed models,

showed the ANN model was better than the MLR model for prediction of methyl orange

removal using Au-NP-AC and Tamarisk. Using the optimal ANN model the coefficient of

determination (R2) for the test data set were 0.958 and 0.989; mean squared error (MSE) values

were 0.00082 and 0.0006 for Au-NP-AC and Tamarisk adsorbent, respectively. In this study a

novel and green approach were reported for the synthesis of gold nanoparticle and activated

carbon by tamarisk. This material was characterized using different techniques such as SEM,

TEM, XRD and BET. The usability of Au-NP-AC and activated carbon (AC) tamarisk for the

methyl orange from aqueous solutions was investigated. The effect of variables such as pH,

initial dye concentration, adsorbent dosage (g) and contact time (min) on methyl orange removal

were studied. Fitting the experimental equilibrium data to various isotherm models such as

Langmuir, Freundlich, Tempkin and Dubinin- Radushkevich models show the suitability and

1
applicability of the Langmuir model. Kinetic models such as pseudo -first order, pseudo-second

order, Elovich and intraparticle diffusion models indicate that the second-order equation and

intraparticle diffusion models control the kinetic of the adsorption process. The small amount of

proposed Au-NP-AC and activated carbon (0.015 g and 0.75 g) is applicable for successful

removal of methyl orange (>98%) in short time (20 min for Au-NP-AC and 45 min for tamarisk-

AC) with high adsorption capacity 161 mg g−1 for Au-NP-AC and 3.84 mg g-1 for tamarisk-AC.

Keywords: Gold nanoparticles, Tamarisk, Methyl orange, Artificial neural network, Partial

swarm optimization.

* Author to whom correspondence should be addressed. Tel & fax: (0098)-741-2223048,

E-mail: m_ghaedi@mail.yu.ac.ir

** Author to whom correspondence should be addressed. Tel: +98-742-3332033, Fax: +98-742-

3332003, E-mail: abm_ghaedi@yahoo.com

2
1. Introduction

As we know, perform experiments for problem solving in engineering are normally costly and

time-consuming. Therefore, we can analyze experimental data to discover many scientific

relationships and apply computational modeling to solve complex engineering problems.

Modeling is a simplification of reality and well-accepted engineering technique. Computational

modeling assists us to analyze and figure out what is going on in a process and confirm results

achieved from experiments. The selection of appropriate procedures for constructing the models

is a main factor to build an accurate prediction. Often used statistical methods such as simple, but

interpretable multiple linear regression (MLR) [1] and nonlinear, highly predictive methods

include fuzzy inference system (FIS) [2], adaptive Neuro-fuzzy inference system (ANFIS) [3,4],

support vector machine (SVM) [5,6], artificial neural networks (ANN) [7-12], and random forest

[13]. Unlike classical optimization algorithm, the meta-heuristic optimization algorithms are

efficient and flexible models. Different heuristic optimization algorithms have been used to solve

the reliability redundancy allocation problem. Sometimes, some impressive ideas are motivated

from biological, physical, and social processing [14, 15]. Particle swarm optimization (PSO) is a

stochastic optimization approach inspired from the simulation of simplified social behavior of

bird flocking, firstly introduced by Kennedy and Eberhart [16]. Khajeh et al., used response

surface methodology (RSM) and combine of artificial neural network-particle swarm

optimization (ANN-PSO) for prediction and optimization of removal of methylene blue using

silver nanoparticles from water samples. Their results indicated that the modeling capability of

the ANN was better than RSM with comparative less value of RMSE and more the value of

correlation coefficient [17]. Khajeh et al. proposed response surface methodology (RSM) and

combine of artificial neural network-particle swarm optimization (ANN-PSO) to develop

predictive models for simulation and optimization of tea waste extraction process. They have
3
shown that the result of the ANN was better than the RSM model [18]. Dyes are one of the most

hazardous materials in industrial effluents, which can cause severe health problems in human

beings, since they exhibit high biotoxicity and potentially mutagenic and carcinogenic effects

[19, 20]. Therefore, the removal of dye from colored effluents has attracted increasing attention.

Several technologies including biological treatment, adsorption, coagulation, flocculation,

chemical oxidation, membrane separation and ion exchange have been developed [21–23].

Among these methods, adsorption has been considered to be simple, highly efficient, and ease of

operation. A wide range of materials has been reported for dye removal, including activated

carbon, zeolite, clay, polymer, etc. [24–26]. In view of pollutant control at present, it is still

indispensable for the development of new adsorbent materials with high adsorption capacities

and removal efficiencies. Methyl Orange is widely used for coloration of cellulosic textiles. Some

of the applied techniques for the treatment of dye contaminated wastewaters are flocculation,

coagulation, precipitation, adsorption, membrane filtration, electrochemical techniques,

ozonation and fungal decolorization [27, 28]. Among them, the adsorption based procedure

widely utilized due to its high efficiency, capacity and ability for large scale dye removal

application (with potential for adsorbent regeneration) [29–33]. The non toxic, low cost and easy

availability adsorbents are the best choice for wastewater treatment. Due to the unique properties

of nanoparticles such as a high number of reactive atoms, high mechanical and thermal strength,

large number of vacant reactive surface sites metallic or semi-metallic behavior and high surface

area widely applied for removal of various toxic materials [34]. Activated carbon as one of the

most common building blocks of nanotechnology is the low cost and general material as support

for loading nanomaterial. In such activated carbon based adsorbents the various reactive sites

such as AC functional group and nanoparticle material synergies ally adsorb trace amount of

different pollutant. Due to its unique molecular structure, AC has an extremely high affinity for
4
many classes of dyes and controlled by the AC characteristics. In this technique application of

nanoscale materials with high surface area enhance the removal percentage and adsorption

capacity of AC based adsorbent. Metallic nanoparticles such as gold as well as gold are of great

importance in the analytical chemistry, catalytic applications, electromagnetic devices, etc. In this

investigation, MLR and ANN models have been used to study linear and nonlinear relationships

exist among variables while PSO used to optimize the influence of variables concern to the

adsorption of methyl orange by Au-NP-AC and AC. The results achieved from the introduced

models were compared with the experimental values. The Au-NP-AC was synthesized and

subsequently characterized via various techniques such as scanning electron microscopy (SEM)

and X-ray diffraction (XRD) analysis. Then the adsorption kinetics and isotherms of methyl

orange removal on this adsorbent was studied and its applicability for treatment of wastewater

and this dye removal was investigated. The adsorption rates were evaluated by fitting the

experimental data to traditional kinetic models such as pseudo first and second-order and

intraparticle diffusion models. The proposed sorbent is useful for quantitative adsorption of the

methyl orange with high sorption capacities in short time.

2. Experimental

2.1. Instruments and Reagents

The stock solution (1000 mg L−1) of Methyl Orange (formula weight: 327.34 g mol−1 and

molecular formula: C14H14N3NaO3S) was prepared by dissolving appropriate amount of the solid

dye in double distilled water and the working concentrations daily were prepared by its diluting

(Fig. 1). The pH measurements were carried out using Metrohm 692 pH/Ion meter (Metrohm,

Switzerland, Swiss) and the methyl orange concentration was determined using Jasco UV-Vis

spectrophotometer model V-530 (Jasco, Japan) at a wavelength of 465 nm. The morphology of
5
the Au-NP-AC and AC was observed by scanning electron microscopy (SEM; Hitachi S-4160,

Japan) under an acceleration voltage of 15 kV. X-ray diffraction (XRD) pattern was recorded by

an automated Philips X'Pert X-ray diffractometer (Philips, Netherland) (40 kV and 30 mA) for 2θ

values over 10-80º. A BET surface analyzer (Quantachrome NOVA 2000, USA) was used to

measure nitrogen adsorption–desorption isotherm at 77 K while before the measurement, the

samples were degassed using helium at 553 K for 3 h. The BET surface area, total pore volume,

and micropore area were obtained from the adsorption isotherms. All chemicals, including

methyl orange, NaOH, HCl and KCl with the highest purity available were purchased from

Merck (Darmstadt, Germany).

2.2. Measurements of dye uptake

The dye concentrations were determined according to generally traditional photometry method at

a maximum wavelength over working concentration. The efficiency of methyl orange removal

was determined at different time intervals (in the range of 2-30 min for Au-NP-AC and 15-60

min for AC-Tamarisk) and the equilibrium was established after 30 min for Au-NP-AC and 45

min for AC-Tamarisk of contact time. The effect of initial pH on the removal of methyl orange

was evaluated in the pH range of 1–9, by contacting 50 mL of 15 mg L−1 of initial dye

concentration with 0.015 g of Au-NP-AC and 14 mg L−1 of initial dye concentration with 0.5 g

AC for 30 and 60 min of contact time. The experiments were also performed in the initial dye

concentration range of 8–18 mg L−1 for Ac-tamarisk and 11-20 mg L-1 for Au-NP-AC to obtain

adsorption isotherms. The methyl orange removal percentage was calculated using the following

equation:

% methyl orange removal = ((C0 – Ct)/C0) × 100 (1)

6
where C0 (mg L−1) and Ct (mg L−1) is the concentration of target at initial and after time t

respectively. The adsorbed methyl orange amount (qe (mg g−1)) was calculated by the following

mass balance relationship:

qe = (C0 − Ce) V/W (2)

where C0 and Ce (mg L−1) are the initial and equilibrium dye concentrations in aqueous solution,

respectively, V (L) is the volume of the solution and W (g) is the mass of the adsorbent.

2.3. Definition ANN model

Artificial neural network models have been widely used to achieve relationships in complex

nonlinear data sets. ANN models have been successfully achieved acceptance in numerous

environmental engineering problems. ANN models have been mainly included of three layers of

an input layer, hidden layer and output layers (Fig. 2). The ANN model contains of simple

processing units called neurons. ANN model is generally a parallel interconnected structure,

including the input layer of neurons, a number of hidden layers, and output layer. The structure of

an ANN model is determined by the number of its layers, the number of nodes in each layer and

the nature of transfer function [1, 35, 36]. In this investigation, Matlab R2011a Neural Network

Toolbox has been utilized to predict the adsorption efficiency. A three layer ANN with a tangent

sigmoid transfer function (tansig) at hidden layer, a linear transfer function (purelin) at output

layer and Levenberg-Marquardt back-propagation algorithm with 1000 iterations were applied.

The data were randomly divided into two groups (70 % for training, and 30% of the testing).

Therefore 182, and 78 samples were implemented for training, and testing subsets, respectively.

Four neurons (concentration (mg L−1), adsorbent dosage (g), contact time (min) and stirrer speed

(rpm)), 1 to 30 neurons in the hidden layer and one neuron (removal percentage) in the output

layer were used. The ANN inputs and outputs are normalized between 0.1 and 0.9 to avoid
7
numerical overflows due to very large or small weights. The normalization equation applied is as

follows:

(3)

where y is the normalized value of xi, the xmax and xmin are the maximum and minimum value of

xi, respectively. The performance of the ANN and MLR models were done according to the mean

squared error (MSE) and the coefficient of determination (R2) which can be defined as follows:

(4)

(5)

where yprd,i was the predicted value by ANN model, yexp,i was the experimental value, N was the

number of data, and ym was the average of the experimental value.

2.4. Multiple linear regression (MLR)

Multiple linear regression (MLR) model is commonly used in environmental studies in order to

model the influence of numerous independent variables on the dependent variable [1]. MLR

model can be defined by the following mathematical equation:

(6)

where Y was the predicted value by MLR model, (i = 0,…,n) are the regression coefficients,

(i = 1,…,n) are the independent variables (inputs). Besides the different purposes of MLR

model, it can be easily used in the adsorption process to determine the cumulative influence of

several independent variables, such as time, adsorbent dosage, and concentration dye of the

system, on a dependent variable (removal percentage). In the study, SPSS 15 statistical program

has been applied to MLR model.

8
2.5. Particle swarm optimization (PSO)

PSO has indicated to be an efficient, powerful and simple optimization algorithm. The PSO
contains various control parameters, such as the acceleration coefficients, inertia weight, velocity
clamping, and swarm size [37, 38]. The theory and more details PSO can be found in the
literatures [39-41]. Fig. 3 shows the flow chart of the PSO, In brief, the PSO includes several
steps that were shown as follows [39, 42].

1. Generation and initialization of an array of particles with random positions and velocities.

2. Evaluation of objective function for each particle.

3. The objective function values are computed for new positions of each particle. If a better

position is obtained by particle, the p Best value is replaced by the current value.

4. Determines if the particle has found the maximal force in the population. If the new g Best

value is better than previous g Best value, the g Best value is replaced by the current. G Best

value and stored. The result of optimization is vector g Best.

5. Calculation of particles’ new velocity.

6. Update particle’s position by moving towards maximal objective function.

7. Steps (1) and (2) are repeated until the iteration number reaches a predetermined iteration.

The PSO has several control parameters to be set, such as the acceleration coefficients (Personal

Learning Coefficient= c1 and Global Learning Coefficient= c2), inertia weight (w), velocity

clamping, and swarm size (nPop). Wrong initialization of these parameters may lead to divergent

or cyclic behavior [38].

2.6. Preparation of activated carbon by Tamarisk

2 kg of the steam wood of the Tamarisk tree was cutted to small parts and washed twisted

thoroughly and was cooked and heated at 350°C in 1.0 L vessel for 2 h. Subsequently, the wood

9
of the Tamarisk tree poured 2 L of de-ionized water and heated to boil for two more hours to

remove the water soluble phenolic compounds and to avoid from their releases during the

adsorption experiments. Then, the Tamarisk tree wastes were washed with distilled water and

dried at 150 ◦C in an air-supplied oven for 12 h and was grounded thereafter in a disk-mill and

sieved (20-30 mesh) subsequently. The carbonization process was taking place in argon

atmosphere with heating rate of 5 °C/min and final temperature was set as 500 °C and kept for 1

h. The mass was then cooled and washed thoroughly several times with distilled water and dried

and fully characterized with different conventional techniques, such as SEM and BET analysis

Table 1. The point of zero charge pH (pHZPC) was examined via well known pH drift method.

Characterization of prepared adsorbent was presented in Table 1. The gold nanoparticle was

synthesized according to literature [41-43].

3. Results and discussion

3.1. Characterization of adsorbent

The SPR of each nano-scale zero vallent element significations depend to pH, temperature, time

and amount of stabilizing agent [43, 44]. The slight blue shift of the SPR from 540 to 525 nm

with enhance in pH is due to decrease in the size of Au nanoparticles [45]. On the other hand, the

structure of starch and amount of producing Au nanoparticle significantly depend on pH. Change

in the structure of starch leads to significant change in the charge and hydrogen bonding and

finally affect the double role of starch as both stabilizing and reducing agents. The uniform size

distribution of Au-nanoparticles (in the range of 20–60 nm) achieved by FESEM (Fig. 4), has a

good agreement with TEM image (Fig. 4 C) size in the range of (25-55 nm). The SEM

micrograph of Tamarisk-AC (1.57KX magnification) presented in Fig. 4 E. It shows

homogeneous structure and appearance of different pores in the adsorbent structure that make

10
possible it as useful sites for adsorption. Determination of specific surface area by N2/77 K

adsorption isotherms assumed to measure the surface area in microspores within pore sizes of a

material [46]. Specific surface area (SSA) is the accessible area of adsorbent surface per unit

mass of material. The interference by the surrounding phase is especially problematic for the

Bruner-Emmet-Teller (BET) N2 adsorption/desorption isotherm method because the entire

surface is modified by vacuum treatment before N2 adsorption. Table 2 and Fig. 5 A-D, Show

that adsorbent possessed appreciable narrow micro-porosity. The surface area of Au-NP-AC was

found to be 1229.55 m2 g-1. Total pore volume is 0.6905 cm3 g-1 and average pore diameter is less

than 22.4635 Å. Total surface properties of adsorbent are presented in Table 2.

3. 2. Optimization of the ANN model

The optimal structure of the ANN model was attained based on the maximum value of R2 and the

minimum value of the MSE of the training and testing data sets. In networks optimization 1 to 30

of neurons were utilized in the hidden layer. Table 3 reveals the dependence among the number

of neurons, R2 and MSE for selected ANN for both adsorbents. It was found that the ANN

containing 11 and 20 hidden neurons was chosen as the best models, for the removal of methyl

orange by Au-NP-AC and Tamarisk adsorbent, respectively. Using the optimal ANN models the

R2 for the test data set were 0.9580 and 0.9886; MSE values were 0.00082 and 0.0006 for gold

nanoparticles loaded on activated carbon and Tamarisk adsorbent, respectively. Fig. 6 exhibits

the MSE versus the number of epochs for optimal ANN models, as can be seen from the figure

the training was stopped after 6 and 8 epochs for Au-NP-AC and Tamarisk, respectively. Figs. 7

A and B shows the predicted values of normalized removal of data in the training, and testing sets

using the ANN models and their plotting versus experimental data for Au-NP-AC and Tamarisk,

11
respectively. The results display that there is a good agreement between experimental data and

predicted data by ANN models using both adsorbent.

3. 3. MLR Model

Multiple linear regressions of normalized removal values for 182 data in the training set were

achieved using the four input and the following equations of (7) and (8) were obtained for Au-

NP-AC and Tamarisk, respectively:

(7)

(8)

where Y, x1, x2, x3 and x4 are the normalized removal percentage, contact time (min), dye

concentration (mg L−1), stirrer speed (rpm) and adsorbent dosage (g), respectively. Then the

model was applied to removal prediction for the testing set. Figs. 7 C and D indicate the predicate

values of normalized removal of data in the training, and testing sets using the MLR models and

their plotting versus experimental data for both adsorbents. Table 4 compares the results obtained

using the MLR and ANN models for both adsorbents. The results indicate the predicted data with

good agreement for the experimental data are predictable, while the ANN efficiency was better

than the MLR model.

3. 4. PSO optimization

PSO optimization was utilized to optimize the adsorption process using both adsorbents. In this

study, PSO based optimization processes performed by equations that obtained from ANN

models. The method starts with a population of random regimes using adsorbent dosage (g),

contact time (min), stirrer speed (rpm) and initial dye concentration (mg L−1) as the optimization

parameters. The optimum ANN models with 11 and 20 neurons in the hidden layer were used as
12
fitness function for Au-NP-AC and Tamarisk, respectively. The weight and bias values of each

layer, which determined the optimum ANN structure, are as follow.

where, w1 and b1 are the weight and bias of hidden layer and w2 and b2 are the weight and bias

of output layers. The following Equation shows the fitness function for ANN-PSO models that

correlate the inputs with output.

13
 ANN=Purelin(w2*tansig(w1*[x(1);x(2);x(3);x(4)]+b1)+b2) (9)

where, x(1), x(2), x(3) and x(4) represent the inputs. The parameters have been applied for PSO

processes with 4000 maximum number of iterations, swarm size of 8, inertia weight of 0.7298,

inertia weight damping ratio of 1, personal learning coefficient of 1.4692, and global learning

coefficient of 1.4692. Fig. 8 A and B shows the best cost values versus iterations for both

adsorbents. As can be seen, after 13 and 11 iterations the value of best cost reaches to minimize

cost and then remains constant for Au-NP-AC and Tamarisk, respectively. The ANN-PSO

method predicts maximum removal at 0.02 of adsorbent dosage (g), 10 of initial dye

concentration (mg L−1), 20 of stirrer speed (rpm) and 30 of contact time (min), for both

adsorbents.

3.5. Effect of pH

It is known that the solution pH can affect the surface charge of the adsorbent, the degree of

ionization of the different pollutants, the dissociation of functional groups on the active sites of

the adsorbent as well as the structure of the dye molecule [27, 31]. So the solution pH is an

important parameter during the dye adsorption process. Fig. 9 A shows the effect of pH on the

removal of methyl orange by Au-NP-AC at room temperature, contact time of 30 min, adsorbent

dosage of 0.015 g in 50 ml and dye concentration of 15 mg L-1. Fig. 9 A also indicate the impact

of pH on the removal of methyl orange by activated carbon Tamarisk at room temperature,

contact time of 60 min, adsorbent dosage of 0.5 g in 50 ml and dye concentration of 10 mg L-1.

Methyl orange (a water-soluble azo dye), is a well-known carcinogenic organic substance which

is widely used in textile, printing, paper manufacturing, pharmaceutical and food industries but

also in research laboratories as an acid base indicator due to its ability to function as a weak acid.

The effect of initial pH on the methyl orange removal in the pH range of 1-9 with an initial dye
14
concentration of 15 mg L−1, 30 min of contact time at room temperature and adsorbent dose of

0.015 g for Au-NP-AC was studied. For AC 14 mg L−1, 60 min of contact time at room

temperature and adsorbent dose of 0.5 g were applied. The respective results (Fig. 9 A) show

that the initial pH significantly affects the extent of adsorption of methyl orange over the Au-NP-

AC and AC the removal percentage significantly decreased at higher pH values. The mechanisms

of adsorption of methyl orange as anionic dye on Au-NP-AC and AC adsorbent may be the

electrostatic interaction between the protonated positive charge adsorbent and anionic dye and the

chemical reaction through the various reactive atoms of methyl orange dye and the adsorbents. It

was seen that the maximum uptake of the methyl orange was obtained at pH 2.0. Therefore, all

subsequent studies were carried out at pH 2. Lower adsorption of methyl orange (anionic dye) at

high pH is related to the electrostatic repulsion between the dye molecule and anionic surface of

adsorbents and the presence of excess of OH− ions that compete with the anionic form of methyl

orange for the adsorption sites. At pH 2 the H+ ion concentration in the system increased and the

surface of the Au-NP-AC acquires a positive charge by absorbing H+ ion. At low pH, both the

AC functional groups such as OH, COOH, C=O and NH and sulfur atoms of Au and AC will

protonated and adsorbent gets a positive charge [30, 31].

3.6. Effects of adsorbent dose

Adsorbent dose is one of the important factors because it is used to determine capacity of an

adsorbent for initial concentration of adsorbate. By monitoring the amount of adsorbent, it is

possible to study the effects of adsorption process to attain a maximum adsorption capacity of the

adsorbent [30, 31]. To explore the effects of adsorbent dose on the adsorption system of methyl

orange, experiments were conducted by varying amount of adsorbent of 0.005g to 0.02g, at a

constant initial dye concentration of 11- 20 mg L-1 for Au-NP-AC and 0.25 g to 1.0 g, at a
15
constant initial dye concentration of 8-18 mg L-1 for AC. The selection of these doses was based

on the preliminary adsorption test. The percent adsorption increased from about 97% to 99% as

the amount of adsorbent increased from 0.005 to 0.02 g for Au-NP-AC and 70% to 99% as the

amount of adsorbent increased from 0.25 to 1.0 g for AC at an equilibrium contact time (Figs. 9

(B) and (C)). It is obvious that as adsorbent dose increases, the number of available adsorption

sites increases; this in turn increases the uptake of the dye. The reason for the decrease in the

adsorption density with increase in adsorbent dose is instauration of adsorption sites through

adsorption process. Another possible reason may be the intra-particle interaction, for instance,

aggregation, as a result of a high adsorbent dose. Because of such aggregation, total surface area

of adsorbent decreases and diffusion path length increases.

3.7. Effects of contact time

Contact time is one of the important parameters for the assessment of practical application of

adsorption process [30, 31]. Figs. 9 (D) and (E) show the uptake of methyl orange as a function

of contact time for adsorbents. Fig. 9 D displays the influence of contact time on methyl orange

removal at 0.005-0.02 g of Au-NP-AC at room temperature, pH=2 and dye concentration of 15

mg L-1. Fig. E also show the effect of contact time on methyl orange removal at 0.25-1.0 g of

activated carbon Tamarisk at room temperature, pH=2 and dye concentration of 16 mg L-1.

Equilibrium is reached within 18 min for Au-NP-AC and 45 min for AC. Beyond equilibrium

time, adsorption is found to be nearly constant. This is mainly because of the fact that at the

initial stage of adsorption, vacant surface sites are available, once equilibrium is attained; the

remaining vacant sites are difficult to be occupied, probably caused by the repulsive forces

between the molecules on the adsorbents the bulk phase.

16
3.8. Effect of initial dye concentration on removal of methyl orange

The effect of initial dye concentration (11-20 mg L-1 at 0.005-0.02 g of Au-NP-AC and 8-18 mg

L-1 at 0.25-1.0 g of activated carbon Tamarisk) on its removal efficiency onto both adsorbent was

studied. It was seen that the methyl orange removal percentage decreased with an increase in its

initial concentration (Figs. 9 (F) and (G)). Immediate increase in initial dye concentrations due to

lack of available un-occupied sites led to decrease in mass transport and removal efficiency. It

seems that subsequently due to adsorbents saturation, the removal percentage decreased, while

the actual amount of dyes significantly increased [28, 30]. For Au-NP-AC and activated carbon

increasing concentration from 11–20 mg L-1 and 8-18 mg L-1 led to decrease in removal

percentage from about 98.6 to 82.8% for Au-NP-AC and 99.7% to 84.6% for AC. On the other

hand, there was a significant increase in the actual amount of adsorbed dye and adsorption

capacity for both adsorbents, especially Au-NP-AC.

3.9. Adsorption kinetic modeling

The prediction of batch adsorption kinetics is necessary for the design of industrial adsorption

columns. The nature of the adsorption process will depend on physical or chemical characteristics

of the adsorbent system and the system conditions. Four kinetic models pseudo first-order,

pseudo second-order, Elovich and intraparticle diffusion model were considered to investigate the

rate and mechanism of an adsorption process and the requirement (Tables 5-8 for Au-NP-AC and

Tables 9-12 for Tamarisk-AC). Lagergren described the pseudo-first-order [47]. In the first order

kinetic model, according to the equation listed in Tables 5-12 by plotting the values of log (qe−qt)

versus t may give a linear relationship that k1 and qe values can be determined from the slope and

intercept of the obtained line, respectively. If the intercept does not equal to experimental qe

value, it can be considered that the reaction is not likely to be a first-order reaction, even this plot

has a high correlation coefficient [48, 49]. A pseudo second-order model may describe the
17
sorption kinetics [50]. In despite of the first order model, the plot of t/qt versus t for the pseudo-

second-order kinetic model gives a straight line with a high correlation coefficient that k2 and

equilibrium adsorption capacity (qe) were calculated from the intercept and slope of this line,

respectively. The values of R2 and closeness of experimental and theoretical adsorption capacity

(qe) value show the applicability of the second order model to explain and interpret the

experimental data (Tables 5-12). The R2 value for pseudo-second-order kinetic model was found

to be higher (1.000) and the calculated qe value is mainly close to the experimental adsorption

capacity value under different physico-chemical conditions. It is important to note that for a

pseudo-first-order model, the linear correlation coefficient is less than the pseudo-second-order

correlation coefficient. The Elovich equation as another rate equation based on the adsorption

capacity in linear form has been successfully applied for the adsorption of solutes from a liquid

solution [51, 52]. Plot of qt versus ln(t) should yield a linear relationship if the Elovich is

applicable with a slope of (1/β) and an intercept of (1/β) ln (αβ). The Elovich constants obtained

from the slope and the intercept of the straight line are reported in Tables 5-12. The correlation

coefficient is higher than 0.9782 shows the suitability of the model for evaluation of the

adsorption process. The dye sorption is governed either by the liquid phase mass transport rate or

through the intra-particle mass transport rate. The later process possibility is explored using the

intraparticle diffusion model based on diffusive mass transfer that adsorption rate expressed in

terms of the square root of time (t) [53-55]. The values of Kdiff and C were calculated from the

slope and intercept of the plot of qt versus t1/2. C value is related to the thickness of the boundary

layer and Kdiff is the intraparticle diffusion rate constant (mg g−1 min−1/2) were reported in Tables

5-12. The values of qt were found to give the two lines part with values of t1/2 and the rate

constant Kdiff directly evaluated from the slope of the second regression line. The first one of

these lines representing surface adsorption at the beginning of the reaction and the second one is
18
the intraparticle diffusion at the end of the reaction. The first line shows a boundary layer effect

at the initial stage of the adsorption. The second line is the gradual adsorption stage where

intraparticle diffusion is the rate limiting step. In some cases, the last and third lines show the

final equilibrium stage (to slow down of intraparticle diffusion). The values of Kdiff and C were

obtained from the final linear portion and their values are presented in Tables 5-12. Since, the

intraparticle curve did not pass through the origin; one can notice that in addition to the

intraparticle diffusion model another stage such as second order kinetic model act control the

adsorption process. The initial adsorption rates h (mg (g min)-1) can be calculated from the

pseudo-second-order model by the following equation:

h=k2qe2 (10)

It was found that the initial rate of adsorption increases with increasing the initial dye. The

variation of the initial rate of adsorption with the initial dye concentration, which would be

expected due to the increase in driving force at higher concentration.

3.10. Adsorption equilibrium study

To optimize the design of the adsorption of the adsorbate, it is important to establish the most

adsorption mechanism, the surface properties and tendency of adsorbent toward each dye.

Various isotherm equations such as Langmuir, Freundlich, Tempkin and Dubinin- Radushkevich

(D-R) isotherms to discuss the equilibrium characteristics of the adsorption process have been

used [56-61]. The constant parameters of the isotherm equations for this adsorption process and

the correlation coefficient (R2) for all the three isotherms in Tables 13 and 14 are summarized.

Based on the linear form of Langmuir isotherm model [50] (according to Tables 13, 14), the

values of KL (the Langmuir adsorption constant (L. mg-1)) and qm (theoretical maximum

adsorption capacity (mg g-1)) were obtained from the intercept and slope of the plot of Ce/qe vs
19
Ce, respectively. The high correlation coefficient shows the applicability of the Langmuir model

for interpretation of the experimental data over the whole concentration range. The parameters of

Freundlich isotherm model such as KF ((mg g-1)/( mg L-1)1/n) and n (the capacity and intensity of

the adsorption) were calculated from the intercept and slope of the linear plot of ln qe versus ln

Ce, respectively and their values are presented in Table 13, 14. The value of 1/n for Freundlich

isotherm 0.094 and 0.13 shows the high tendency of methyl orange for the adsorption onto Au-

NP-AC and AC, while lower R2 (0.8504 and 0.4524) value shows its unsuitability for fitting the

experimental data over the whole concentration range. The heat of the adsorption and the

adsorbent–adsorbate interaction by using Tempkin isotherm model were evaluated [61]. B is the

Tempkin constant related to heat of the adsorption (J mol−1), T is the absolute temperature (K), R

is the universal gas constant (8.314 J mol−1.K−1), KT is the equilibrium binding constant (L mg−1).

The values of the Tempkin constants and the correlation coefficient are lower than the Langmuir

value. Therefore, the Langmuir isotherm provides the best correlation for the experimental data.

Another adsorption isotherm commonly used for liquid phase adsorption on a surface was

proposed by Dubinin-Radushkevich (D-R). The D–R model to estimate the porosity apparent free

energy and the characteristic of adsorption was applied. In the D–R isotherm K (mol2 (kJ2) −1) is a

constant related to the adsorption energy, qm (mg g−1) is the theoretical saturation capacity, ε is

the Polanyi potential. The slope of the plot of ln qe versus ε2 gives K and the intercept yields the

qm value. The mean free energy of the adsorption (E), defined as the free energy change when

one mole of ion is transferred from the solution to the surface of the sorbent, can be calculated.

3.11. Comparison with other adsorbents for methyl orange

Many methyl orange removal research using various biosorbent and adsorbents were reported in

the literature [62-67] and their performance for the methyl orange removal was compared with
20
term of adsorption capacity. It can be seen that methyl orange removal by Au-NP-AC and AC

Tamarisk (Table 15). Is superior to previously reported literature higher adsorption capacity in a

short time.

4. Conclusion

The AC and Au-NP-AC are identified to be effective adsorbents for the removal of methyl

orange from aqueous solutions. Adsorption parameters for the Langmuir, Freundlich, and

Dubinin–Radushkevich and Freundlich isotherms were determined and the equilibrium data were

best described by the Tempkin isotherm model for application of AC and Langmuir for

application of Au-NP-AC. The adsorption kinetics can be successfully fitted to pseudo-second-

order kinetic model for both adsorbents. The results of the intraparticle diffusion model suggested

that this kinetic model was not the only rate controlling step for methyl orange adsorption on all

adsorbents and also suggests that AC and Au-NP-AC could be employed as promising low-cost a

high capacity adsorbent for removal of methyl orange from aqueous solutions. Linear (MLR) and

nonlinear (ANN) models were applied to simulate adsorption of methyl orange using Au-NP-AC

and Tamarisk. Comparison of the results achieved using proposed models showed three layer

ANN models with 11 and 20 neurons in the hidden layer is a more suitable model to predict the

adsorption of methyl orange using Au-NP-AC and Tamarisk, respectively. AC-Tamarisk and Au-

NP-AC as high surface area adsorbents (high adsorption capacity) need short extraction time

(≤18 min for Au-NP-AC and 45 min for AC-Tamarisk) with low consumption of adsorbent (0.5 g

of AC and 0.015 g of Au-NP-AC). Although Au-NP-AC has the higher adsorption capacity (161

mg g−1) compared to AC (3.84 mg g−1) but since Au-NP-AC is a toxic adsorbent, it is better and

safe to use AC as green and safe adsorbent for removal of methyl orange.

21
References
[1] U. Özdemir, B. Özbay, S. Veli, S. Zor, Chem. Eng. J. 178 (2011) 183-190.
[2] M. Razani, A. Yazdani-Chamzini, S. Haji Yakhchali, Safety Sci. 55 (2013) 26-33.

[3] M. Khalighi, S. Farooq, I. A. Karimi, Comput. Aided. Chem. Eng. 31 (2012) 1336-1340.

[4] M. Ghaedi, A.M. Ghaedi, F. Abdi, M. Roosta, A. Vafaei, A. Asghari, Ecotox. Environ. Safe.
96 (2013) 110-117.
[5] Y. Liu, F. Sun, Expert. Syst. Appl. 40 (2013) 4496-4502.

[6] M. Ghaedi, A.M. Ghaedi, M. Hossainpour, A. Ansari, M.H. Habibi, A.R. Asghari, J. Ind.
Eng. Chem. http://dx.doi.org/10.1016/j.jiec.2013.08.011.
[7] S. Elemen, E. P. A. Kumbasar, S. Yapar, Dyes Pigments, 95 (2012) 102-111.

[8] S. M. Hosseini Asl, M. Ahmadi, M. Ghiasvand, A. Tardast, R. Katal, J. Ind. Eng. Chem. 19
(2013) 1044-1055.

[9] R. Hosseini Nia, M. Ghaedi, A.M. Ghaedi, . Mol. Liq. 195 (2014) 219-229.
[10] M. Ghaedi, N. Zeinali, A.M. Ghaedi, M. Teimuori, J. Tashkhourian, Spectrochim. Acta Part
A, 125 (2014) 264-277.
[11] M. Ghaedi, A.M. Ghaedi, F. Abdi, M. Roosta, R. Sahraei, A. Daneshfar, J. Ind. Eng. Chem.
20 (2014) 787-795.
[12] M. Ghaedi, A.M. Ghaedi, E. Negintaji, A. Ansari, F. Mohammadi, J. Ind. Eng. Chem.
http://dx.doi.org/10.1016/j.jiec.2014.01.041.

22
[13] M. Ghaedi, A.M. Ghaedi, E. Negintaji, A. Ansari, A. Vafaei, M. Rajabi, J. Ind. Eng. Chem.
http://dx.doi.org/10.1016/j.jiec.2013.08.033.
[14] B. Alatas, Expert. Syst. Appl. 39 (2012) 11080–11088.
[15] L. D. Afonso, V. C. Mariani, L. S. Coelho, Expert. Syst. Appl. 40 (2013) 3794–3802.
[16] J. Kennedy, R. Eberhart, Proc. IEEE Int'l. Conf. on Neural Networks, Perth, Australia, 4
(1995) 1942–1948.
[17] M. Khajeh, M. Kaykhaii, A. Sharafi, J. Ind. Eng. Chem. 19 (2013) 1624–1630.
[18] M. Khajeh, A. Sarafraz-Yazdi, A. Fakhrai-Moghadam, Arabian. J. Chem.
http://dx.doi.org/10.1016/j.arabjc.2013.06.011
[19] M. Riera-Torres, C. Gutiérrez-Bouz‫ل‬n, M. Crespi, Desalination 252 (2010) 53–59.
[20] J. Labanda, J. Sabaté, J. Llorens, Chem. Eng. J. 166 (2011) 536–543.
[21] A. Demirbas, J. Hazard. Mater. 167 (2009) 1–9.
[22] L.H. Ai, Y. Zhou, J. Jiang, Desalination 266 (2011) 72–77.
[23] A. Dabrowski, Adv. Colloid Interface Sci. 93 (2001) 135–224.
[24] K.K.H. Choy, J.F. Porter, G. McKay, Chem. Eng. J. 103 (2004) 133–145.
[25] H. E. Bakouria, J. Useroa, J. Morilloa. A. Ouassini, Bioresour. Technol. 100 (2009) 4147–
4155.
[26] N. Hoda, E. Bayram, E Ayranci, J. Hazard. Mater. 137 (2006) 344–351.
[27] M. Ghaedi, H. Khajehsharifi, A. Hemmati Yadkuri, M. Roosta. A. Asghari, Toxicol.
Environ. Chem. 94 (2012) 873–883.
[28] M. Ghaedi, A. Hassanzadeh, S. Nasiri Kokhdan, J. Chem. Eng. 56 (2011) 2511–2520.
[29] A. Duran, M. Tuzen, M. Soylak, J. Hazard. Mater. 169 (2009) 466–471.
[30] M. Ghaedi, A. Ansari, R. Sahraei, Spectrochimica Acta Part A. 114 (2013) 687–694.
[31] M. Ghaedi, A. Ansari, M.H. Habibi, A.R. Asghari, J. Ind. Eng. Chem. 20 (2014) 17-28.
[32] M. Tuzen, M. Soylak, J. Hazard. Mater. 147 (2007) 219–225.
[33] M. Ghaedi, K. Mortazavi, M. Jamshidi, M. Roosta, B. Karami, Toxicol. Environ. Chem. 94
(2012) 846–859.
[34] M. Ghaedi, S. Heidarpour, S. Nasiri Kokhdan, R. Sahraie, A. Daneshfar, B. Brazesh, Powder
Technol. 228 (2012) 18–25.
[35] A. Celekli, S. S. Birecikligil, F. Geyik, H. Bozkurt, Bioresour. Technol. 103 (2012) 64–70.
[36] S. Aber, A.R. Amani-Ghadim, V. Mirzajani, J. Hazard. Mater. 171 (2009) 484–490.

23
[37] J. Kennedy, Proc. IEEE Int. Conf. Evol. Comput., Indianapolis, USA. (1997) 303–308.
[38] F. van den Bergh, A.P. Engelbrecht, Inf. Sci. 176 (2006) 937–971.
[39] N. Yusup, A. M. Zain, S. Z. M. Hashim, Proc. Eng. 29 (2012) 914 – 923.
[40] M. S. Leu, M. F. Yeh, S. C. Wang, Appl. Soft. Comput. 13 (2013) 4047–4062.
[41] A. Khare, S. Rangnekar, Appl. Soft. Comput. 13 (2013) 2997–3006.
[42] U. Zuperl, F. Cus, V. Gecevska, J. Mech. Engin. 6 (2007) 354-368.
[43] G. Karimipour, M. Ghaedi, R. Sahraei, A. Daneshfar, M. Nejati Biyareh, Biol. Trace Elem.
Res. 145 (2012)109-117.
[44] M. Brust, J. Fink, D. Bethell, J. D. Schiffrin, C. Kiely, J. Chem. Soc. Chem. Commun., 16
(1995) 1655-1656.
[45] T.Q. Tuan, N.V. Son, H.T.K. Dung, N.H. Luong, B.T. Thuy, N.T.V. Anh, N.D. Hoa, N.H.
Hai, J. Hazard. Mater. 192 (2011) 1321–1329.
[46] C.Y. Flores, C. Diaz, A. Rubert, G.A. Benitez, M.S. Moreno, M.A. Fernandez, L. de Mele,
R.C. Salvarezza, P.L. Schilardi, C. Vericat, J. Colloid Interface Sci. 350 (2010) 402–408.
[47] S. Lagergren, Kungliga Svenska Vetenskapsakademiens, Handlingar 24 (1898) 1–39.
[48] Y.S. Ho, G. Mckay, D.A.J. Wase, C.F. Foster. Adsorpt. Sci. Technol. 18 (2000) 639–650.
[49] D.M. Ruthven, K.F. Loughlin, Chem. Eng. Sci. 26 (1971) 577–584.
[50] Y.S. Ho, G. McKay, Biochem. 34 (1999) 451–465.
[51] E. Bulut, M. Ozacar, Micropor. Mesopor. Mater. 115 (2008) 234–246.
[52] S.H. Chien, W.R. Clayton, Soil Sci. Soc. Am. J. 44 (1980) 265–268.
[53] J.J. Pignatello, F.J. Ferrandino, L.Q. Huangm, Environ. Sci. Technol. 27 (1993) 1563–1571.
[54] M. Ghaedi, A. Hekmati, S. Khodadoust, R. Sahraei, A. Daneshfar, A. Mihandoost, M.K.
Purkait, Spectrochim. Acta Part A 90 (2012) 22–27
[55] D. Ozdes, C. Duran, H. B.Senturk, J. Environ. Manage. 92 (2011) 3082–3090.
[56] M. Ghaedi, S. Zamani Amirabad, F. Marahel, S. Nasiri Kokhdan, R. Sahraei. A. Daneshfar,
Spectrochim. Acta Part A 83 (2011) 46–51.
[57] I. Langmuir, J. Am. Chem. Soc. 38 (1916) 2221–2295.
[58] H.M.F. Freundlich, J. Phys. Chem. 57 (1906) 385–470.
[59] C. Ng, J.N. Losso, W.E. Marshall, R.M. Rao, Bioresoure. Technol. 85 (2002) 131–135.
[60] L. H. Aia, C. Zhang, Z. Chen, J. Hazard. Mater. 192 (2011) 1515– 1524.
[61] M.J. Temkin, V. Pyzhev, Acta Physiochim. USSR 12 (1940) 217–222.

24
[62] G. Annadurai, R.S. Juang, D.J. Lee, J. Hazard. Mater. 92 (2002) 263–274.
[63] N. Mohammadi, H. Khani, V.K. Gupta, E. Amereh, S. Agarwal, J. Colloid. Interface. Sci.
362 (2011) 457–462.
[64] R. Liu, K. Fu, B. Zhang, D. Mei, H. Zhang, J. Liu, J. Dispersion. Sci. Technol. 33 (2012)
711–718.
[65] G. Li, Y. Du, Y. Tao, H. Deng, X. Luo, J. Yang, Carbohydr. Polym. 82 (2010) 706–713.
[66] H.Y. Zhu, R. Jiang, L. Xiao, G.M. Zeng, Bioresource. Technol. 101(2010) 5063-5069.
[67] S. Hosseini, M.A. Khan, M.R. Malekbala, W. Cheah, T.S.Y. Choong, Chem. Eng. J.
171(2011) 1124-1131.
[68] J. Pal, M. Kanti Deb, D. Kumar Deshmukh, D. Verma, Appl. Water. Sci. 3 (2013) 367–374

25
 Table 1: Characteristics of prepared Tamarisk-AC

Property Values

Elements presents (%) C: 76; O: 22; Al: 0.5; Cl: 1.5

BET specific surface area (m2/g) 8.798

Alkali soluble 1.0 wt %

Water soluble 0.3%

1.0 wt %
Acid soluble

pHZPC 5.5

26
 Table 2 Summary Report of Au-NP-AC properties

Summary Report
Surface Area
BET Surface Area: 1229.5535 m²/g
BJH Adsorption cumulative surface area of pores
between 17.000 Å and 3000.000 Å width: 288.625 m²/g
BJH Desorption cumulative surface area of pores
between 17.000 Å and 3000.000 Å width: 307.6435 m²/g
Pore Volume
BJH Adsorption cumulative volume of pores 0.266307 cm³/g
between 17.000 Å and 3000.000 Å width:
BJH Desorption cumulative volume of pores 0.272894 cm³/g
between 17.000 Å and 3000.000 Å width:
Pore Size
Adsorption average pore width (4V/A by BET): 22.4635 Å
BJH Adsorption average pore width (4V/A): 36.907 Å
BJH Desorption average pore width (4V/A): 35.482 Å
Nanoparticle Size
Average Particle Size : 48.798 Å

27
 Table 3 Dependence among neuron number at hidden layer, MSE and R2
Gold nanoparticles Tamarisk
Number of Train Test Train Test
neurons MSE R2 MSE R2 MSE R2 MSE R2
1 0.0095 0.5327 0.0082 0.5809 0.0024 0.9423 0.0019 0.9630
2 0.0032 0.8408 0.0025 0.8728 0.0029 0.9297 0.0030 0.9430
3 0.0018 0.9116 0.0019 0.9024 8.68e-004 0.9789 7.57e-004 0.9855
4 0.0035 0.8299 0.0022 0.8872 0.0013 0.9686 0.0012 0.9764
5 0.0036 0.8210 0.0052 0.7321 0.0011 0.9724 0.0021 0.9592
6 0.0056 0.7239 0.0053 0.7313 0.0011 0.9744 7.89e-004 0.9849
7 0.0018 0.9111 0.0027 0.8617 7.34e-004 0.9822 6.71e-004 0.9872
8 0.0012 0.9387 8.571e-004 0.9562 5.58e-004 0.9865 7.63e-004 0.9854
9 0.0037 0.8166 0.0035 0.8190 5.82e-004 0.9859 7.94e-004 0.9848
10 0.0020 0.9005 0.0013 0.9351 0.0010 0.9754 0.0015 0.9716
11 0.0012 0.9420 8.224e-004 0.9580 7.77e-004 0.9811 0.0022 0.9584
12 0.0013 0.9366 0.0023 0.8818 4.19e-004 0.9898 7.78e-004 0.9851
13 0.0013 0.9365 0.0012 0.9378 6.74e-004 0.9836 0.0016 0.9689
14 0.0017 0.9162 0.0023 0.8828 6.79e-004 0.9835 0.0012 0.9768
15 0.0033 0.8376 0.0033 0.8340 0.0014 0.9654 0.0023 0.9553
20 0.0010 0.9492 0.0016 0.9197 4.38e-004 0.9893 5.98e-004 0.9886
25 0.0011 0.9476 0.0025 0.8736 6.38e-004 0.9845 0.0011 0.9786
30 0.0014 0.9334 0.0021 0.8905 0.0013 0.9683 0.0016 0.9691

28
Table 4 Comparison of MSE and R2 obtained using the MLR and ANN models for both
adsorbents
Gold nanoparticles Tamarisk
Model Training set Testing set Training set Testing set
2 2 2
MSE R MSE R MSE R MSE R2
ANN 0.0012 0.9420 0.00082 0.9580 0.00044 0.9893 0.00060 0.9886
MLR 0.0082 0.5973 0.0071 0.6379 0.011 0.732 0.0123 0.782

29
Table 5 Kinetic parameters of methyl orange adsorption onto Au-NP-AC Conditions: 0.005 g
adsorbent over 11-20 mg L-1 at optima conditions of other variables

Parameter values: Concentration dye (ppm)

Models parameters 11 13 15 17 19 20
First order kinetic model: K1 0.1564 0.152 0.125 0.073 0.09 0.079

Log (qe-qt)=log(qe)-(K1/2.303)t
qe (cal) 21.232 27.479 24.119 27.16 52.24 65.02

R2 0.9805 0.9487 0.9202 0.947 0.9506 0.7941

Second order kinetic model: K2 0.0176 0.0132 0.0098 0.012 0.0057 0.0044

t/qt=1/k2qe2 + (1/qe)t
qe (cal) 109.89 129.87 149.25 161.3 175.4 151.51

R2 0.9998 0.9997 0.9991 0.9991 0.999 0.9884

h 212.77 200.0 222.22 312.52 175.48 103.12

Intraparticle diffusion Kdif 4.6784 6.4327 6.64 5.475 4.89 12.473

qt=Kid t1/2+C
C 87.674 99.579 116.14 133.56 1.718 88.737
2
R 0.9177 0.8614 0.9807 0.9531 0.9506 0.8808

Elovich β 0.1522 15.174 0.111 0.135 0.056 0.058

qt= 1/β ln(αβ) + 1/β ln (t) α 9.8*105 2.5*105 4.4*106 6.1*108 1.3*104 3196

R2 0.9572 0.9487 0.9501 0.9279 0.9457 0.8898

Experimental date qe(exp) 108.32 127.86 148.37 163.28 177.29 155.11

30
Table 6 Kinetic parameters of methyl orange adsorption onto Au-NP-AC Conditions: 0.01 g
adsorbent over 11-20 mg L-1 at optima conditions of other variables

Parameter values: Concentration dye (ppm)

Models parameters 11 13 15 17 19 20
First order kinetic model: K1 0.1329 0.117 0.102 0.082 0.103 0.056

Log (qe-qt)=log(qe)-(K1/2.303)t
qe (cal) 6.99 7.454 7.023 9.31 24.8 23.53

R2 0.9753 0.9675 0.9696 0.985 0.9007 0.6576

Second order kinetic model: K2 0.0534 0.0486 0.0507 0.036 0.0124 0.0136

t/qt=1/k2qe2 + (1/qe)t
qe (cal) 54.94 64.102 74.074 81.96 89.28 78.125

R2 0.9999 0.9998 0.9999 0.9998 0.9975 0.9899

h 161.33 200.0 277.85 243.91 99 83.49

Intraparticle diffusion Kdif 1.524 1.5977 1.6147 1.1445 4.89 3.379

qt=Kid t1/2+C
C 47.556 56.53 66.336 49.341 64.215 59.226
2
R 0.9628 0.9571 0.9149 0.9855 0.9604 0.6259

Elovich β 0.469 0.4502 0.44 0.382 0.149 0.233

10 11 13 12 5
qt= 1/β ln(αβ) + 1/β ln (t) α 1.2*10 2.8*10 1.3*10 3.4*10 1.2*10 6.2*106

R2 0.9935 0.9784 0.9579 0.9458 0.946 0.7346

Experimental date qe(exp) 54.58 64.19 74.23 82.53 89.34 80.11

31
Table 7 Kinetic parameters of methyl orange adsorption onto Au-NP-AC Conditions: 0.015 g
adsorbent over 11-20 mg L-1 at optima conditions of other variables

Parameter values: Concentration dye (ppm)

Models parameters 11 13 15 17 19 20
First order kinetic model: K1 0.112 0.07 0.092 0.069 0.084 0.098

Log (qe-qt)=log(qe)-(K1/2.303)t
qe (cal) 2.3238 2.303 4.570 5.66 4.59 9.08

R2 0.976 0.9388 0.967 96.72 0.9462 0.9602

Second order kinetic model: K2 0.1609 0.1712 0.078 0.061 0.224 0.00363

t.qt=1/k2qe2 + (1/qe)t
qe (cal) 36.36 42.735 49.5 54.94 57.47 60.97

R2 1 1 0.9999 0.9998 0.9999 0.9998

h 212.78 312.72 191.191 185.178 740.3 135.127

Intraparticle diffusion Kdif 0.5564 0.5301 1.08 1.1445 1.069 0.0424

qt=Kid t1.2+C
C 33.793 40.185 44.359 49.341 52.39 .958

R2 0.9248 0.9103 0.9047 0.9551 0.9286 0.9602

Elovich β 1.2709 1.329 0.655 0.652 0.683 0.337

qt= 1/β ln(αβ) + 1/β ln (t) α 3.7*1018 1.3*1023 4.5*1012 3.3*1027 6.6*1015 8.8*107

R2 0.9782 0.9711 0.9573 0.9053 0.9216 0.9371

Experimental date qe(exp) 36.41 43.04 49.73 55.87 57.87 61.01

32
Table 8 Kinetic parameters of methyl orange adsorption onto Au-NP-AC Conditions: 0.02 g
adsorbent over 11-20 mg L-1 at optima conditions of other variables

Parameter values: Concentration dye (ppm)

Models parameters 11 13 15 17 19 20
First order kinetic model: K1
Log (qe-qt)=log(qe)-(K1/2.303)t 0.1322 0.282 0.0948 0.0946 0.108 0.131

qe (cal) 1.459 1.326 2.7 3.18 9.436 8.69

R2 0.9735 0.8379 0.9088 0.9482 0.9063 0.8438

Second order kinetic model: K2

t/qt=1/k2qe2 + (1/qe)t 0.266 0.3144 0.139 0.117 0.062 0.00424

qe (cal) 27.397 32.051 37.17 42.019 45.05 47.39

R2 1 1 1 1 0.9995 0.9993

h 200.05 323.02 192.425 208.43 125.05 135.127

Intraparticle diffusion Kdif

qt=Kid t1/2+C 0.3547 0.0297 0.7608 0.8345 0.9914 1.5412

C 25.757 0.1227 33.738 38.289 39.894 39.58

R2 0.9235 0.8379 0.8082 0.8438 0.9516 0.9784

Elovich β
qt= 1/β ln(αβ) + 1/β ln (t) 1.991 2.116 0.899 0.975 0.770 0.484

α 1022 5.9*1027 1.9*1013 2*1016 3.9*1013 5.1*108

Experimental date R2 0.9791 0.8994 0.9131 0.9314 0.8637 0.929

qe(exp) 27.35 32.3 37.32 42.28 45.06 47.19

33
Table 9 Kinetic parameters of methyl orange adsorption onto Tamarisk-AC Conditions: 0.25 g
adsorbent over8-18 mg L-1 at optima conditions of other variables

Parameter values: Concentration dye (ppm)

Models parameters 8 10 12 14 16 18
First order kinetic model: K1
0.0274 0.0209 0.0345 0.0294 0.0232 0.0193
Log (qe-qt)=log(qe)-(K1/2.303)t
qe (cal) 1.301 1.411 1.261 1.14 1.244 1.657
2
R 0.8043 0.923 0.9471 0.9809 0.9645 0.9157
Second order kinetic model: K2
0.0741 0.090 0.036 0.045 0.446 0.0284
t/qt=1/k2qe2 + (1/qe)t
qe (cal) 1.301 1.386 1.982 2.048 2.198 2.313
2
R 0.9566 0.9867 0.9798 0.9936 0.9954 0.952
h 0.125 0.173 0.143 0.1897 0.2156 0.146
Intraparticle diffusion Kdif
0.0805 0.0784 0.1504 0.0128 0.1554 0.1794
qt=Kid t1/2+C
C 0.5138 0.6496 0.5077 0.0576 0.7597 0.5088
2
R 0.7935 0.9254 0.5077 0.9809 0.9671 0.9039
Elovich β 4.852 4.892 2.53 2.611 2.394 2.143
qt= 1/β ln(αβ) + 1/β ln (t) α 734 1.393 0.344 1.6 0.682 0.547
2
R 0.7278 0.8799 0.933 0.9684 0.9776 0.8555

Experimental date qe(exp) 1.284 1.45 1.79 1.96 2.23 2.42

Table 10 Kinetic parameters of methyl orange adsorption onto Tamarisk -AC Conditions: 0.5 g
adsorbent over 8-18 mg L-1 at optima conditions of other variables

34
 Parameter values: Concentration dye (ppm)

Models parameters 8 10 12 14 16 18
First order kinetic model: K1
0.0516 0.0325 0.0297 0.031 0.0486 0.0196
Log (qe-qt)=log(qe)-(K1/2.303)t
qe (cal) 5.513 5.486 2.888 10.063 1.376 3.64
2
R 0.9643 0.8962 0.6757 0.8426 0.8031 0.5142
Second order kinetic model: K2
0.543 0.469 0.2456 0.317 0.1138 0.308
t/qt=1/k2qe2 + (1/qe)t
qe (cal) 0.816 0.993 1.168 1.361 1.564 1.48
R2 0.9997 0.999 0.9927 0.9937 0.9912 0.9971
h 0.362 0.463 0.335 32.1 0.2805 0.676
Intraparticle diffusion Kdif
1/2 0.0201 0.0213 0.0366 0.0492 0.0697 0.0453
qt=Kid t +C
C 0.6455 0.8051 0.8382 0.9319 0.9214 1.1191
2
R 0.9785 0.9496 0.8099 0.8649 0.9036 0.508
Elovich β 18.63 17.89 10.47 7.886 5.509 7.739
qt= 1/β ln(αβ) + 1/β ln (t) α 2380 2.8*1014 174.407 57.106 8.278 173.07
2
R 0.9816 0.9174 0.7724 0.8046 0.8568 0.5776

Experimental date qe(exp) 0.79 0.99 1.18 1.38 1.48 1.52

Table 11 Kinetic parameters of methyl orange adsorption onto Tamarisk-AC Conditions: 0.75 g
adsorbent over 8-18 mg L-1 at optima conditions of other variables

35
 Parameter values: Concentration dye (ppm)

Models parameters 8 10 12 14 16 18
First order kinetic model: K1
0.0592 0.03 0.031 0.0354 0.0152 0.02
Log (qe-qt)=log(qe)-(K1/2.303)t
qe (cal) 7.168 11.084 9.618 8.742 6.237 4.72
2
R 0.9034 0.9319 0.9772 0.8657 0.7962 0.9092
Second order kinetic model: K2
0.809 0.977 0.840 0.672 0.639 0.435
t.qt=1/k2qe2 + (1/qe)t
qe (cal) 0.547 0.662 0.793 0.918 0.994 1.130
2
R 0.9996 0.9996 0.9998 0.9996 0.9986 0.9989
h 0.243 0.428 0.528 0.566 0.632 0.555
Intraparticle diffusion Kdif
1.2 0.0131 0.0109 0.0129 0.0132 0.015 0.0234
qt=Kid t +C
C 0.4337 0.5675 0.6815 0.8031 0.8588 0.9243
2
R 0.9937 0.9723 0.9868 0.9465 0.8192 0.9385
Elovich β 28.57 34.72 29.24 64.93 25.77 16.23
qt= 1/β ln(αβ) + 1/β ln (t)
α 4664 2.8*106 4.2*106 3.6*1026 4.5*107 5.6*104
2
R 0.9875 0.9497 0.9673 0.8657 0.7676 0.9118

Experimental date qe(exp) 0.53 0.66 0.79 0.91 1.04 1.17

36
Table 12 Kinetic parameters of methyl orange adsorption onto Tamarisk-AC Conditions: 1 g
adsorbent over 8-18 mg L-1 at optima conditions of other variables

Parameter values: Concentration dye (ppm)

Models parameters 8 10 12 14 16 18
First order kinetic model: K1
0.0446 0.692 0.034 0.04 0.050 0.0242
Log (qe-qt)=log(qe)-(K1/2.303)t
qe (cal) 23.89 20 21.557 13.93 7.112 9.523
2
R 0.847 0.8851 0.913 0.8263 0.9534 0.9319
Second order kinetic model: K2
2.413 1.852 1.992 1.319 0.709 0.845
t/qt=1/k2qe2 + (1/qe)t
qe (cal) 0.402 0.501 0.596 0.698 0.7888 0.853
2
R 0.9997 0.9997 0.9998 0.9997 0.9998 0.9997
h 0.390 0.465 0.7076 0.643 0.441 0.636
Intraparticle diffusion Kdif
1/2 0.0043 0.0056 0.0054 0.0085 0.0156 0.0128
qt=Kid t +C
C 0.3647 0.4519 0.5489 0.6246 0.6551 0.7426
2
R 0.9242 0.9215 0.947 0.9263 0.9871 0.9486
Elovich β
90.09 69.44 71.43 44.64 23.98 29.33
qt= 1/β ln(αβ) + 1/β ln (t)

α 2.3*1014 1.73*1011 4.2*1014 7.8*109 7.4*104 2.6*107

2
Experimental date R 0.8753 0.8732 0.9072 0.909 0.987 0.9431
qe(exp) 0.4 0.5 0.59 0.69 0.77 0.86

37
Table 13 isotherm constants of methyl orange adsorption onto Au-NP-AC

Isotherm Equation parameters Adsorbent (g)

0.005 0.01 0.015 0.02
-1
Langmuir Ce/qe = 1/KaQm + Ce/Qm Q m (mg g ) 161.29 84.74 60.60 47.61
-1
Ka (L mg ) 0.0129 59 33 35
2
R 0.9979 0.9959 0.9986 0.9986
Freundlich Ln qe = ln KF +(1/n)ln Ce 1/n 0.0941 0.1108 0.1089 0.1182
KF (L mg -1) 155.13 81.69 57.57 46.46
2
R 0.4524 0.682 0.6614 0.5954
Tempkin qe = BT ln KT + BT lnCe BT 13.314 7.9293 5.4004 4.5081
5 4 4
KT (L mg ) -1
1.3*10 3.1*10 4.3*10 2.9*104
R2 0.4655 0.7019 0.712 0.6481
2 -1
Dubinin and Ln qe = ln Qs – Kε Qs (mg g ) 171.48 88.99 61.28 48.15
Radushkevich
K 10-8 10-8 8*10-9 7*10-9
E (kJ/mol)= 3.54 7.07 7.91 8.46
1/(2K)1/2
R2 0.6062 0.8941 0.7425 0.6602

38
Table 14 isotherm constants of methyl orange adsorption onto Tamarisk-AC

Isotherm Equation Parameters Adsorbent (g)

0.25 0.5 0.75 1.0

Langmuir Ce/qe = 1/KaQm + Ce/Qm Q m (mg g-1) 3.84 1.176 1.54 0.863
-1
Ka (L mg ) 39.96 29.22 28.73 0.273
2
R 0.9999 0.9977 0.9944 0.9977
Freundlich Ln qe = ln KF +(1/n)ln Ce 1/n 0.1344 0.5019 0.1572 0.2074
-1
KF (L mg ) 1.454 1.018 1.111 1.302
R2 0.8504 0.9442 0.969 0.9752
Tempkin qe = Bl ln KT + Bl lnCe Bl 0.1555 0.8913 0.0949 0.1667
KT (L mg -1) 11.49 2.363 0.9136 1.480
2
R 0.8702 0.9284 0.9682 0.9561
2 -1
Dubinin and Ln qe = ln Qs – Kε Qs (mg g ) 1.501 2.306 1.241 1.114
Radushkevich
B 7*10-9 4*10-7 5*10-9 7*10-9
E (kJ/mol)= 2.702 1.122 8.474 8.333
1/2
1/(2K)
R2 0.9411 0.7984 0.9423 0.9502

39
Table 15 Comparison of the adsorption capacity values calculated of methyl orange adsorption

reported in the literatures and the present work

Adsorption Source
Adsorbent Capacity (mg g-1)
Banana peel 21 [62]
Orange peel 20.5 [62]
Mesoporous carbon material 294.1 [63]
Modified halloysite 94.34 [64]
nanotubes
Iron(II) cross-linked chitin- 107.5 [65]
based gel beads
Magnetic chitosan 66.09 [66]
Carbon coated monolith 102.04 [67]
AgNPs-coated AC 27.48 [68]
Au-NP-AC 161 This work
AC-Tamarisk 3.84 This work

40
Figure captions

Fig. 1. Chemical structure of methyl orange changes by pH.

Fig. 2. Structure of a back-propagation artificial neural network

Fig. 3. Flow chart of the PSO

Fig. 4. A) The UV-Vis absorption spectra of the Au nanoparticles stabilized in a 0.15 wt %

aqueous starch dispersion under different pH conditions. B) FESEM image of the Au
nanoparticles loaded onto activated carbon. C) Typical TEM image of the starch-stabilized Au
nanoparticles. D) The electron diffraction (ED) pattern of the Au nanoparticles. E) SEM images
of the Ac-Tamarisk
Fig.5. A) BJH Adsorption Cumulative Pore Volume (Larger) Halsey : Faas Correction of Au-
NP-AC. B) BJH Adsorption Cumulative Pore Area (Larger) Halsey : Faas Correction of Au-NP-
AC. C) Isotherm Linear Plot of Au-NP-AC. D) Isotherm Linear Absolute Plot of Au-NP-AC.
Fig. 6. The MSE versus the number of epochs for optimal ANN models, A) Au-NP-AC and B)
Tamarisk

Fig. 7. The Predicted values of normalized removal versus experimental data for training, and
testing sets using the ANN model A) Au-NP-AC, B) Tamarisk, and using the MLR models, C)
Au-NP-AC, D) Tamarisk

Fig. 8. The best cost values against iteration, A) Au-NP-AC and B) Tamarisk

41
Fig. 9. Effect of pH, adsorbent dosage, contact time and initial dye concentration on the removal
of methyl orange A, B, D and F, respectively, by Au-NP-AC and A, C, E and G, respectively by
Tamarisk.

H3C OH- H3C

+ - -
N N NH SO3 N N N SO3

H3C H+ H3C

Fig. 1.

42
Fig. 2.

43
Fig. 3

44
45
Fig 4.

46
Fig. 5.

47
Fig. 6. A

Fig. 6 B

48
Fig. 7 A

Fig. 7 B

49
Fig. 7 C

Fig. 7 D

50
Fig. 8 A

Fig. 8B

51
Fig. 9 A

Fig. 9 B

52
Fig. 9 C

Fig. 9 D

53
Fig. 9 E

Fig. 9 F

54
Fig. 9 G

55
56
Dear Editor

A novel nanoparticle has been used as reusable support for removal of dye EY.

A high efficiency removal technology is proposed for fast removal of dye.

Removal of dye can be significantly enhanced under application of nanoparticle.

The adsorption process was modeled by artificial neural network.

This model is applicable for rapid removal of large quantity of the dye.

Sincerely yours,

M. Ghaedi

57
58