artificial neural network approach for modeling of adsorption of ni (ii) and cr (vi) ions...
DESCRIPTION
Rapid industrialization is seriously contributing to the release of heavy metals into the main water bodies. Increasing concentration of toxic heavy metals in the water sources constitutes severe health hazard due to their toxicity. Adsorption is one of the methods in effective removal of metal ions from water. Adsorption isotherms are one of the ways of expressing equilibrium relationship between adsorbate and adsorbent however very little has been reported in the literature related to the equilibrium relationship between two sorbets and two adsorbents in one isotherm. Estimation of adsorption isotherms which are adsorbent & adsorbate combination specific, become prerequisite in designing adsorption processes. The conventional mathematical model fails very often in developing correlation because of complexity and nonlinearity governing this process. Artificial Neural Network is the black box modeling tool that can address to the modeling of the operations involving multivariable nonlinear relationship and also can incorporate linguistic variables in coded number form. In present work Artificial neural network has been applied for adsorption of heavy metals Ni (II) & Cr (VI) simultaneously present in aqueous solution using the synthesized adsorbents. Adsorption studies are performed for aqueous solutions in the metal ion concentration range of 0.4938 to 1.2345 mg/ml for Ni (II) and 0.1944 to 0.4418 mg/ml for Cr (VI) with adsorbent dosing of 1-5 gm. Three ANN models ACS, ACM & ACC with different topology have been developed using elite-ANN with sigmoid function for estimation of % adsorption, equilibrium concentration and amount of adsorbent adsorbed per unit adsorbent .These estimated values for all the parameters are compared for their prediction accuracy for both the training and test data sets. The results are indicative that the ANN model ACM with two hidden layers & 10 neurons in each hidden layer has high accuracy compared to other two models. The novel feature of this work is in highlighting the potential of ANN models in substituting adsorption isotherms that will enable the user to incorporate linguistic variables coded with numbers for multiple adsorbents & adsorbates into a single model with ease & high accuracy.TRANSCRIPT
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
114 Vol. 6, Issue 1, pp. 114-127
ARTIFICIAL NEURAL NETWORK APPROACH FOR
MODELING OF ADSORPTION OF NI (II) AND CR (VI) IONS
SIMULTANEOUSLY PRESENT IN AQUEOUS SOLUTION USING
ADSORBENT SYNTHESIZED FROM AEGEL MARMELOS FRUIT
SHELL AND SYZYGIUM CUMINI SEED
S. L. Pandharipande1, Aarti R. Deshmukh2
1Associate Professor, 2M. Tech Third Semester,
Department of Chemical Engineering, Laxminarayan Institute of Technology, Rashtrasant
Tukadoji Maharaj Nagpur University, Bharat Nagar, Amravati Road, Nagpur, India.
ABSTRACT
Rapid industrialization is seriously contributing to the release of heavy metals into the main water bodies.
Increasing concentration of toxic heavy metals in the water sources constitutes severe health hazard due to their
toxicity. Adsorption is one of the methods in effective removal of metal ions from water. Adsorption isotherms
are one of the ways of expressing equilibrium relationship between adsorbate and adsorbent however very little
has been reported in the literature related to the equilibrium relationship between two sorbets and two
adsorbents in one isotherm. Estimation of adsorption isotherms which are adsorbent & adsorbate combination
specific, become prerequisite in designing adsorption processes. The conventional mathematical model fails
very often in developing correlation because of complexity and nonlinearity governing this process. Artificial
Neural Network is the black box modeling tool that can address to the modeling of the operations involving
multivariable nonlinear relationship and also can incorporate linguistic variables in coded number form. In
present work Artificial neural network has been applied for adsorption of heavy metals Ni (II) & Cr (VI)
simultaneously present in aqueous solution using the synthesized adsorbents. Adsorption studies are performed
for aqueous solutions in the metal ion concentration range of 0.4938 to 1.2345 mg/ml for Ni (II) and 0.1944 to
0.4418 mg/ml for Cr (VI) with adsorbent dosing of 1-5 gm. Three ANN models ACS, ACM & ACC with different
topology have been developed using elite-ANN with sigmoid function for estimation of % adsorption,
equilibrium concentration and amount of adsorbent adsorbed per unit adsorbent .These estimated values for all
the parameters are compared for their prediction accuracy for both the training and test data sets. The results
are indicative that the ANN model ACM with two hidden layers & 10 neurons in each hidden layer has high
accuracy compared to other two models. The novel feature of this work is in highlighting the potential of ANN
models in substituting adsorption isotherms that will enable the user to incorporate linguistic variables coded
with numbers for multiple adsorbents & adsorbates into a single model with ease & high accuracy.
KEYWORDS: Artificial Neural Network, modeling, heavy metal ion, adsorption, aegel marmelos, syzygium
cumini.
I. INTRODUCTION
The presence of heavy metals in water is a major concern due to their adverse effect on human health.
The discharge of waste water containing heavy metals in the water bodies is thus worrying for
toxicological reasons. Industries such as steel, metal plating, mining, textile, explosive manufacturing,
ceramic & glass, leather, paint, etc., are some of the sources for heavy metal effluents. The heavy
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
115 Vol. 6, Issue 1, pp. 114-127
metals are nonbiodegradable and cannot be recovered economically from the contaminated water.
There are several methods such as coagulation, ion exchange, membrane technology, floatation,
adsorption that are available for removal of heavy metals from waste water. With increasing
environmental awareness on discharge of waste water, there is a need of cost effective methods.
Adsorption is one of the promising techniques in removal of heavy metals from waste water; in terms
of its high adsorption capacity, wide range of target pollutants & its simplicity of design. Hence
researchers are making interest towards the synthesis of low cost adsorbents from agricultural and
industrial wastes which are being good alternatives for the commercial adsorbents
1.1 Adsorption isotherm Adsorption isotherm is one of the possible ways of representation of equilibrium relationship that is
governing the phenomenon of adsorption. Most of the work related reported in literature is devoted
for the determination of the equilibrium concentration of adsorbate which is distributed between
liquid & solid phases, since there could be a very large number of combinations of adsorbates &
adsorbents possible, hence such a type of work becomes relevant. This equilibrium information is
essential in design & estimation of adsorption process.
II. ARTIFICIAL NEURAL NETWORK
Artificial Neural Network is inspired by the working principle of natural networks of biological
neurons. The basic processing element of a neural network is called a neuron or node. The neuron
impulse or the output of a node is calculated as weighted sum of the input signals from the proceeding
neuron, altered by the transfer function. The learning capability of a neuron is accomplished by
adjusting the weights in conformity to chosen learning algorithm. The process is iterative for artificial
neural network.
The basic ANN architecture consists of three types of layers input, hidden & output layers. Number of
neurons in input and output layer depends upon the number of input & output parameters respectively.
The selection of the number of neuron in a hidden layer is an important decision however there is no
definite formula.
There are several types of architecture of ANN. However Multilayer Perceptron (MLP) observed to
be effective in modeling of chemical processes. MLP trained by the back propagation algorithm is
based on a system capable of modeling complex relationship between the variables. ANN is the
powerful tool for modeling, especially when the data relationship is unknown. ANN can identify and
learn correlated patterns between input data set and corresponding output data set. After training ANN
can be used to predict the output of the new independent input data [1, 2].
There are number of applications of ANN, that include, standardization of digital colorimeter [3],
estimation concentration heavy metals from aqueous solution[4], estimation of composition of a
ternary liquid mixture [5], mass transfer predictions in a fast fluidized bed of fine solids [6], modeling
for estimation of hydrodynamics of packed column [7], fault diagnosis in complex chemical plants
[8], adsorption studies [9,10,11,12], modeling combined VLE of four quaternary mixtures [13] and
similar other[ 14,15,16] are also reported.
The present work aims at developing Artificial Neural Network model in adsorption studies for the
removal of Ni (II) and Cr (VI) simultaneously present in aqueous solution and ANN model for
comparison of the % adsorption, equilibrium concentration and the amount of adsorbate adsorbed per
unit amount of adsorbent.
The paper is presented in different sections that include the materials used and methods adopted in
synthesizing the adsorbents from aegel marmelos fruit shell and syzygium cumini seed. A section is
devoted for explaining in detail various feature of Artificial Neural network modeling which includes
the details of architecture & topology, results and discussion in graphical form comparing the actual
and predicted values for all the output parameters. The paper concludes with the inference drawn on
the suitability of the neural network in modeling adsorption parameters for simultaneous removal of
two sorbets on two different sorbents.
III. MATERIAL AND METHODS 3.1 Material for adsorbent
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
116 Vol. 6, Issue 1, pp. 114-127
The adsorbents Aegel Marmelos adsorbent (AMA) & Syzygium Cumini adsorbent (SCA) are
synthesized by thermal &/or chemical methods of Aegel Marmelos fruit shell & Syzygium Cumini
seed [17]. These are used in the adsorption studies of the present work.
3.2 Methodology The first part of the present work is related to the adsorption experimentation that includes removal of
two heavy metal ions Ni (II) and Cr (VI) simultaneously present in aqueous solution using two
adsorbents synthesized from Aegel marmalos & Syzygium cumini seed. Known volume of adsorbate
is added to a known amount of adsorbent and once the equilibrium is reached its pH and optical
density is measured by standardized pH meter and colorimeter. Analysis of sample is done using pH
and optical density [4]. The experimental values of equilibrium concentration (mg/ml), amount of
adsorbate adsorbed per unit amount of adsorbent (mg/gm) and % adsorption are calculated for each
adsorbate with variable amount of adsorbent as given in Table 2. The second part is devoted for
development of artificial neural network models for estimation of equilibrium concentration of heavy
metals, amount of adsorbate adsorbed per unit amount of adsorbent and their % adsorption.
3.3 Developing ANN model The accuracy of the ANN model is dependent upon number of factors that include selection of input
parameters, the number of hidden layers & number of neurons in each hidden layer among others.
One of the most important factors in the ANN model is the determination of the number of hidden
layers to be used.
In present work, elite-ANN © [18] is used in developing numerous combinations of neural network
topology varying with one to three hidden layers so as to arrive at optimal model.
There are four input parameters, initial concentration of Ni (II) and Cr (VI), adsorbent coding, for two
types of adsorbent, adsorbent dosing for quantity of adsorbent added. These are correlated with six
output parameters that include equilibrium concentration of Ni (II) & Cr (VI), amount of adsorbate
adsorbed per unit amount of adsorbent of Ni (II) & Cr (VI), % adsorption of Ni (II) & Cr (VI)
respectively.
The type of adsorbent is a linguistic term which is coded with a number, like, 10 for aegel marmelos
fruit shell adsorbent (AMA) and 20 for syzygium cumini seed adsorbent (SCA).
The total data set of 21 points is given in Table 1. The first 17 data points are used as training data set
and the remaining 4 data points as test data set.
The details of the neural network architecture for the selected model ACM is shown in Figure 1. The
snapshot of elite-ANN© in run mode & the error versus iteration during training mode are shown in
Figures 2 and 3 respectively.
Three ANN models ACS, ACM & ACC that have been developed, are as given in Table 1.
The comparison between RMSE values for training & test data sets for all the models developed has
been carried out. The ANN model ACM is selected based on this criterion of lower RMSE.
Steps involved in simulation run using elite-ANN:
Depending on the input and output parameters, specify the number of input and output
neurons in input and output layers.
Select the complexity level. It gives the information about the number of hidden layers in the
model. In the given software there are three complexity levels simple with two hidden layers
having 5 neurons each, complex with two hidden layers having 10 neurons each and moderate
having three hidden layers with 10 neurons each.
Select the total number of iterations to be performed to train the neural network.
Training the network with training data set.
Testing the train neural network model for its accuracy with test data set.
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
117 Vol. 6, Issue 1, pp. 114-127
Figure 1. Neural Network Architecture for model ACM
Table 1. Neural network topology for ANN models
Model
code
Number of Neurons Data points RMSE
Input
layer
1st
hidden
layer
2nd
hidden
layer
3rd
hidden
layer
Out
put
layer
Training
data set
Test
data
set
Training
data set
Test data
set
ACS 4 00 05 05 6 17 4 0.02371 0.1941
ACM 4 00 10 10 6 17 4 0.00252 0.1206
ACC 4 10 10 10 6 17 4 0.00348 0.1520
Number of iterations = 50000
Input parameters: Initial concentration Ni(II) and Cr (VI), adsorbent coding, adsorbent dosing
Output parameters: Equilibrium concentration of Ni (II) & Cr (VI), amount of adsorbate adsorbed per unit
amount of adsorbent of Ni (II) & Cr (VI), % adsorption of Ni (II) & Cr (VI) respectively.
Table 2. Total data for ANN modeling
Sr.
No
Initial
concentration
(mg/ml)
Type
of
adsor
bent
Amt.
of
adsor
bent
(gm)
Equilibrium
concentration
(mg/ml)
Amount of
adsorbate
adsorbed per unit
amount of
adsorbent, qe
(mg/gm)
Percentage
adsorption
Ni (II) Cr (VI) Ni (II) Cr (VI) Ni (II) Cr (VI) Ni (II) Cr (VI)
1 1.2345 0.4418 AMA 1 0.0492 0.3342 59.26 5.374 96.02 24.33
2 1.2345 0.4418 AMA 3 0.2631 0.1883 16.19 4.224 78.69 57.38
3 1.2345 0.4418 AMA 5 0.0754 0.2816 11.59 1.601 93.89 36.24
4 0.4938 0.1944 AMA 3 0.2333 0.1 4.342 1.573 52.75 48.56
5 0.4938 0.1944 AMA 5 0.2206 0.0943 2.732 1.001 55.33 51.48
6 0.8024 0.3136 AMA 1 0.3081 0.1815 24.72 6.603 61.6 42.12
7 0.8024 0.3136 AMA 5 0.2795 0.126 5.229 1.876 65.17 59.81
8 0.9876 0.3799 AMA 1 0.0498 0.3537 46.89 1.309 94.95 6.89
9 0.9876 0.3799 AMA 3 0.0962 0.2704 14.86 1.825 90.26 28.83
10 0.9876 0.3799 AMA 5 0.0937 0.2734 8.939 1.064 90.52 28.02
11 1.2345 0.4418 SCA 1 0.2631 0.1883 48.57 12.67 78.69 57.38
12 1.2345 0.4418 SCA 5 0.2714 0.1254 9.631 3.163 78.01 71.61
13 0.6173 0.2209 SCA 1 0.2191 0.0938 19.9 6.356 64.49 57.54
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
118 Vol. 6, Issue 1, pp. 114-127
14 0.6173 0.2209 SCA 3 0.2175 0.0931 6.662 2.13 64.75 57.85
15 0.6173 0.2209 SCA 5 0.2148 0.0917 4.024 1.292 65.19 58.48
16 0.8641 0.3092 SCA 1 0.2909 0.138 28.66 8.548 66.33 55.29
17 0.8641 0.3092 SCA 3 0.2516 0.1112 10.21 3.3 70.89 64.04
18 0.4938 0.1944 AMA 1 0.2313 0.0992 13.12 4.759 53.16 48.97
19 0.8024 0.3136 AMA 3 0.2945 0.1411 8.465 2.876 63.3 55.02
20 1.2345 0.4418 SCA 3 0.292 0.1551 15.71 4.778 76.34 64.89
21 0.8641 0.3092 SCA 5 0.2367 0.1025 6.274 2.067 72.61 66.84
Table 3. Actual and Predicted values for the out parameters obtained from ANN model ACM
Sr.
No.
Equilibrium concentration
(mg/ml)
Amount of adsorbate
adsorbed per unit amount of
adsorbent, qe (mg/gm)
Percentage adsorption
Actual Predicted Actual Predicted Actual Predicted
1st 2nd 1st 2nd 3rd 4th 3rd 4th 5th 6th 5th 6th
Ni (II) Cr
(VI)
Ni (II) Cr
(VI)
Ni
(II)
Cr
(VI)
Ni
(II)
Cr
(VI)
Ni
(II)
Cr
(VI)
Ni
(II)
Cr
(VI)
1 0.0492 0.3342 0.05 0.334 59.26 5.374 58.87 5.37 96.02 24.33 95.72 24.34
2 0.2631 0.1883 0.263 0.188 16.19 4.224 16.18 4.22 78.69 57.38 78.68 57.39
3 0.0754 0.2816 0.075 0.281 11.59 1.601 11.56 1.6 93.89 36.24 93.88 36.23
4 0.2333 0.1000 0.233 0.099 4.342 1.573 4.274 1.56 52.75 48.56 53.03 48.55
5 0.2206 0.0943 0.22 0.094 2.732 1.001 3.276 1.14 55.33 51.48 55.25 51.46
6 0.3081 0.1815 0.307 0.181 24.72 6.603 24.7 6.59 61.6 42.12 61.57 42.09
7 0.2795 0.126 0.279 0.126 5.229 1.876 5.143 1.85 65.17 59.81 65.2 59.76
8 0.0498 0.3537 0.051 0.352 46.89 1.309 46.89 1.3 94.95 6.89 95.04 7.025
9 0.0962 0.2704 0.096 0.27 14.86 1.825 14.87 1.82 90.26 28.83 90.25 28.79
10 0.0937 0.2734 0.093 0.274 8.939 1.064 8.974 1.12 90.52 28.02 90.53 27.88
11 0.2631 0.1883 0.262 0.188 48.57 12.67 48.57 12.6 78.69 57.38 78.71 57.29
12 0.2714 0.1254 0.27 0.125 9.631 3.163 9.647 3.13 78.01 71.61 78.11 71.2
13 0.2191 0.0938 0.219 0.0928 19.9 6.356 19.88 6.35 64.49 57.54 64.45 57.53
14 0.2175 0.0931 0.217 0.093 6.662 2.13 6.66 2.13 64.75 57.85 64.72 57.85
15 0.2148 0.0917 0.215 0.093 4.024 1.292 3.911 1.22 65.19 58.48 65.17 58.46
16 0.2909 0.138 0.291 0.138 28.66 8.548 28.65 8.54 66.33 55.29 66.29 55.28
17 0.2516 0.1112 0.251 0.111 10.21 3.3 10.19 3.29 70.89 64.04 70.87 64.06
18 0.2313 0.0992 0.2512 0.1045 13.12 4.759 7.024 2.879 53.16 48.97 53.18 45.18
19 0.2945 0.1411 0.2987 0.0959 8.465 2.876 4.041 6.058 63.3 55.02 54.59 59.54
20 0.292 0.1551 0.2935 0.1403 15.71 4.778 25.9 7.939 76.34 64.89 70.57 71.09
21 0.2367 0.1025 0.2569 0.0968 6.274 2.067 5.874 1.491 72.61 66.84 71.28 67.19
Table 3 gives the actual values obtained from the experimentation and predicted values obtained from
ANN model ACM.
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
119 Vol. 6, Issue 1, pp. 114-127
Figure 2. Snapshot of Simulation run
Figure 3. Iterations verses RMSE for training & test data sets during training
IV. RESULTS AND DISCUSSION
The model ACM developed that has been shortlisted is used for prediction of output parameters for
given set of input parameters for both the training & test data sets.
Figures 4 & 5 and 6 & 7 depict the comparison of actual and predicted values of equilibrium
concentration of Ni (II) and Cr (VI) for training & test data sets respectively as obtained by
ANN model ACM. The nature of the graphs depicts in these figures indicates high level of
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
120 Vol. 6, Issue 1, pp. 114-127
accuracy for prediction of Ni (II) & Cr (VI) equilibrium concentration. Similarly, Figures 8 &
9 and 10 & 11 depict the comparison of actual and predicted values of qe i.e. amount of
adsorbate adsorbed per unit amount of adsorbent (mg/gm) and Figures 12 & 13 and 14 & 15
depicts comparison of actual and predicted values of the percentage adsorption of Ni (II) and
Cr (VI) respectively for both training & test data sets.
Figure 4. Comparison of actual and predicted values of equilibrium concentration of Ni (II) for training data set
obtained by model ACM
Figure 5. Comparison of actual and predicted values of equilibrium concentration of Ni (II) for test data set
obtained by model ACM
Figure 6. Comparison of actual and predicted values of equilibrium concentration of Cr (VI) for training data
set obtained by model ACM
Figure 7. Comparison of actual and predicted values of equilibrium concentration of Cr (VI) for test data set
obtained by model ACM
0
0.1
0.2
0.3
0.4
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
1thActualoutput
1thPredicted output
Data points
eq
ulib
riu
mco
nc.
o
f N
i
0
0.1
0.2
0.3
0.4
1 2 3 4
1thActualoutput1thPredictedoutput
Data points
eq
ulib
riu
mco
nc.
o
f N
i
0
0.1
0.2
0.3
0.4
1 2 3 4 5 6 7 8 9 1011121314151617
2thActualoutput
2thPredicted output
Data points
eq
ulib
riu
mco
nc.
of
Cr
0
0.05
0.1
0.15
0.2
1 2 3 4
2th Actualoutput
2thPredictedoutput
Data points
eq
ulib
riu
mco
nc.
o
f C
r
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
121 Vol. 6, Issue 1, pp. 114-127
Figure 8. Comparison of actual and predicted values qe of Ni (II) for training data set obtained by model ACM
Figure 9. Comparison of actual and predicted values qe of Ni (II) for test data set obtained by model ACM
Figure 10. Comparison of actual and predicted values qe of Cr (VI) for training data set obtained by model
ACM
Figure 11. Comparison of actual and predicted values qe of Cr (VI) for test data set obtained by model ACM
Figure 12. Comparison of actual and predicted values of % adsorption of Ni (II) for training data set obtained
by model ACM
0
20
40
60
80
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
3thActualoutput
3thPredicted output
Data pointsq
e N
i
0
10
20
30
1 2 3 4
3th Actualoutput
3thPredictedoutput
Data points
qe
N
i
0
5
10
15
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
4thPredictedoutput
4th Actualoutput
Data points
qe
C
r
0
5
10
1 2 3 4
4th Actualoutput
4thPredictedoutput
Data points
qe
C
r
0
50
100
150
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
5thActualoutput
5thPredictedoutput
Data points
% a
dso
rpti
on
Ni
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
122 Vol. 6, Issue 1, pp. 114-127
Figure 13. Comparison of actual and predicted values of % adsorption of Ni (II) for test data set obtained by
model ACM
Figure 14. Comparison of actual and predicted values of % adsorption of Cr (VI) for training data set obtained
by model ACM
Figure 15. Comparison of actual and predicted values of % adsorption of Cr (VI) for test data set obtained by
model ACM
The accuracy claims of ACM are further substantiated by calculation of % relative error for
each data point and is depicted in figure 16 & 17 and 18 & 19 and figure 20 & 21 and figure
22 & 23 and figure 24 & 25 and 26 & 27 for training and test data set for equilibrium
concentrations, qe and % adsorption of Ni (II) and Cr (VI) respectively.
The range of distribution of % relative error for the output parameters of the training & test
data sets has been carried out and given in Table 4.
As can be seen from the Table 4, that the % relative error for most of the data points is within
±10 which is indicative of the success of the ACM model developed.
Table 4. Distribution of % relative error for data points for ANN model ACM
Output parameters
Metal
ion
% Relative error
Training data point = 17 Test data points = 04
0- ±05 ±05-±10 >±10 0- ±05 ±05-±10 >±10
Equilibrium concentration Ni (II) 17 00 00 02 02 00
Cr (VI) 17 00 00 01 02 01
Amount of adsorbate adsorbed
per unit amount of adsorbent (qe)
Ni (II) 16 00 01 00 01 03
Cr (VI) 14 02 01 00 00 04
% adsorption Ni (II) 17 00 00 02 01 01
Cr (VI) 17 00 00 01 03 00
% Relative error = (Actual value –Predicted value)/ Actual value × 100
0
50
100
1 2 3 4
5th Actualoutput
5thPredictedoutput
Data points% a
dso
rpti
on
Ni
0
20
40
60
80
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
6th Actualoutput
6thPredictedoutput
Data points
% a
dso
rpti
on
Cr
0
50
100
1 2 3 4
6thActualoutput6thPredictedoutput
Data points
% a
dso
rpti
on
Cr
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
123 Vol. 6, Issue 1, pp. 114-127
Figure 16. % Relative error for equilibrium concentration of Ni (II) for training data set obtained by model
ACM
Figurev17. % Relative error for equilibrium concentration of Ni (II) for test data set obtained by model ACM
Figure 18. % Relative error for equilibrium concentration of Cr (VI) for training data set obtained by model
ACM
Figure 19. % Relative error for equilibrium concentration of Cr (VI) for test data set obtained by model ACM
Figure 20. % Relative error for qe of Ni (II) for training data set obtained by model ACM
-3
-2
-1
0
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Data points
% r
ela
tive
err
or
-10
-5
0
1 2 3 4
Data points
% r
ela
tive
err
or
-1.5
-1
-0.5
0
0.5
1
1.5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Data points
% r
ela
tive
err
or
-20
0
20
40
1 2 3 4
Data points
% r
ela
tive
err
or
-30
-20
-10
0
10
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Data points
% r
ela
tive
err
or
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
124 Vol. 6, Issue 1, pp. 114-127
Figure 21. % Relative error for qe of Ni (II) for test data set obtained by model ACM
Figure 22. % Relative error for qe of Cr (VI) for training data set obtained by model ACM
Figure 23. % Relative error for qe of Cr (VI) for test data set obtained by model ACM
Figure 24. % Relative error for % adsorption of Ni (II) for training data set obtained by model ACM
Figure 25. % Relative error for % adsorption of Ni (II) for training data set obtained by model ACM
-100
-50
0
50
100
1 2 3 4
Data points
% r
ela
tive
err
or
-15
-10
-5
0
5
10
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Data points
% r
ela
tive
err
or
-150
-100
-50
0
50
1 2 3 4
Data points
% r
ela
tive
err
or
-0.6
-0.4
-0.2
0
0.2
0.4
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Data points
% r
ela
tive
err
or
-5
0
5
10
15
1 2 3 4
Data points
% r
ela
tive
err
or
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
125 Vol. 6, Issue 1, pp. 114-127
Figure 26. % Relative error for % adsorption of Cr (VI) for training data set obtained by model ACM
Figure 27. % Relative error for % adsorption of Cr (VI) for test data set obtained by model ACM
V. CONCLUSION
The present work addresses to the modeling of adsorption of Ni(II) and Cr(VI) ions simultaneously
present in aqueous solution onto two adsorbents synthesized from aegel marmelos fruit shell and
syzygium cumini seed using artificial neural network.
Three ANN models ACS, ACM & ACC with different topology have been developed for estimation
of % adsorption, equilibrium concentration of Ni(II) & Cr(VI) present in aqueous solution and amount
of adsorbate adsorbed per amount of adsorbent as a function of initial concentration of Ni(II) and
Cr(VI), adsorbent dosage and coded number for adsorbent type. Based on the RMSE values 0.0252
and 0.1206 for training and test data sets respectively & with 4-10-10-6 architecture the ANN model
ACM has found to be the best amongst the three models.
It can be concluded that the ANN model developed has excellent accuracy and can be effectively used
for adsorption processes involving two sorbets and two sorbents.
The unique feature of the ANN model developed is that it can substitute the conventional adsorption
isotherms that enable the user to incorporate linguistic variables coded with numbers for multiple
adsorbents and multiple adsorbates into a single model with ease & high accuracy. The work is
demonstrative & can be expanded and extended to several such adsorption processes.
VI. FUTURE SCOPE
There are several possible combinations of metallic ions present in waste water and to be removed
from it. The present work can be extended to several such adsorption situations. Involving multiple
adsorbates simultaneously adsorbed from the aqueous solution. ANN application can be further
investigated for representation of adsorption isotherm for this system involving multiple adsorbates.
ACKNOWLEDGEMENT
Authors are thankful to Director, LIT, Nagpur for the facilities and encouragement provided.
REFERENCES
[1]. Anderson J.A (1999) An Introduction to Neural Networks, Prentice-Hall of India, Pvt Ltd New
Delhi.
[2]. Rumelhart D E & McClelland (1986) Back Propagation Training Algorithm Processing, M.I.T
Press, Cambridge Massachusetts.
-3
-2
-1
0
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Data points
% r
ela
tive
err
or
-15
-10
-5
0
5
10
1 2 3 4
Data points
% r
ela
tive
err
or
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
126 Vol. 6, Issue 1, pp. 114-127
[3]. R. D. Khonde & S. L. Pandharipande, (2011) “Application of Artificial Neural Network for
Standardization of Digital Colorimeter”, International Journal of Computer Applications, ICCIA-
5, pp1-4.
[4]. S.L. Pandharipande, Aarti R. Deshmukh and Rohit Kalnake, (2013) “ Artificial Neural Network
modelling for estimation of concentration of Ni (II) and Cr (VI) present in aqueous solution”,
International Journal of Advances in Engineering & Technology (IJAET), Vol. 5, Issue 2, pp122-
131.
[5]. S. L. Pandharipande, Anish M. Shah & Heena Tabassum, (2012) “Artificial Neural Network
Modeling for Estimation of Composition of a Ternary Liquid Mixture with its Physical Properties
such as Refractive Index, pH and Conductivity”, International Journal of Computer Applications,
Vol. 45, No. 9, pp26-29.
[6]. Zamankhan, P., Malinen, P., Lepomaki, H., (1997) “Application of Neural Networks to Mass
Transfer Predictions in a Fast Fluidized Bed of Fine Solids”, AIChE, Vol. 43, pp1684-1690.
[7]. S. L. Pandharipande & Ankit Singh (2012) “Optimizing topology in developing artificial neural
network model for estimation of hydrodynamics of packed column”, International Journal of
Computer Applications, Vol. 58, No. 3, pp49-53.
[8]. J. C. Hoskins , K. M. Kaliyur & David M. Himmelblau, (1991) “Fault diagnosis in complex
chemical plants using artificial neural networks”, AIChE, Vol.37, No.1,pp137-141.
[9]. Kaan Yetilmezsoy, Sevgi Demirel, (2008) “Artificial neural network (ANN) approach for
modeling of Pb(II) adsorption from aqueous solution by Antep pistachio (Pistacia Vera L.)
shells”, Journal of Hazardous Materials,Vol.153, pp1288-1300.
[10]. R. D. Khonde & S. L. Pandharipande, (2012) “Artificial Neural Network modeling for adsorption
of dyes from aqueous solution using rice husk carbon”, International Journal of Computer
Application, Vol. 41, No.4, pp1-5.
[11]. S L Pandharipande, Y.D. Urunkar, Ankit Singh,(2012) “Comparative Study of Topology of ANN
Models for Adsorption of Colouring Agents from Aqueous Solutions using Adsorbents
Synthesized from Agricultural Waste Material”. IJAERS, Vol. I, pp214-216
[12]. N. Gamze Turan, Basac Mesci, Okan Ozgonenel, (2011) “Artificial neural network (ANN)
approach for modeling Zn(II) adsorption from Leachate using a new biosorbent”, Chemical
Engineering Journal 173, pp98– 105.
[13]. Shekhar Pandharipande & Anish M Shah, (2012) “Modeling combined VLE of four quaternary
mixtures using artificial neural network”, International Journal of Advances in Engineering,
Science and Technology (IJAEST), Vol. 2, No. 2, pp169-177.
[14]. S. L. Pandharipande , Aditya Akheramka, Ankit Singh & Anish Shah, (2012) “Artificial Neural
Network Modeling of Properties of Crude Fractions with its TBP and Source of Origin and
Time”, International Journal of Computer Application, Vol. 52, No.15, pp20-25.
[15]. S A Mandavgane, S L Pandharipande & D Subramanian, (2006) “Modeling of desilication of
green liquor using Artificial Neural Network”, International journal of chemical technology, Vol.
13, pp168-172.
[16]. H.R. Godini, M. Ghadrdan, M.R. Omidkhah & S.S. Madaeni, (2011) “Part II: Prediction of the
dialysis process performance using Artificial Neural Network (ANN)”, Desalination, Vol. 265,
pp11-21.
[17]. S. L. Pandharipande, Aarti R. Deshmukh, (2012) “Synthesis, characterization and adsorption
studies for adsorbent synthesized from aegel marmelos shell for removal of Cr (VI) from aqueous
solution”, International Journal of Advanced Engineering Research and Studies (IJAERS), Vol. 2,
Issue 2, pp95-96.
[18]. Pandharipande S L & Badhe Y P, elite-ANN©, ROC No SW-1471/2004.
[19]. O. Olayinka Kehinde, T. Adetunde Oluwatoyin, and O. Oyeyiola Aderonke, (2009) “Comparative
analysis of the efficiencies of two low cost adsorbents in the removal of Cr(VI) and Ni(II) from
aqueous solution, African Journal of Environmental Science and Technology, Vol. 3 (11),
November, pp360-369.
[20]. K.G. Sreejalekshmi, K. Anoop Krishnan, T.S. Anirudhan, (2009) “Adsorption of Pb(II) and
Pb(II)-citric acid on sawdust activated carbon: Kinetic and equilibrium isotherm studies”, Journal
of Hazardous Materials 161, pp1506–1513.
[21]. M. Ahmaruzzaman, (2011) “Industrial wastes as low-cost potential adsorbents for the treatment of
wastewater laden with heavy metals”, Advances in Colloid and Interface Science 166 pp36-59.
[22]. Menderes Koyuncu, (2012) “Adsorption of Cr (VI) from textile waste water by using natural
bentonite”, Institute of Integrative Omics and Applied Biotechnology Journal, Vol. 3, pp1-4.
[23]. Hülya Genç-Fuhrman, Peng Wu, Yushan Zhou, Anna Ledin, (2008), “Removal of As, Cd, Cr, Cu,
Ni and Zn from polluted water using an iron based sorbent”, Desalination 226, pp357–370
International Journal of Advances in Engineering & Technology, Mar. 2013.
©IJAET ISSN: 2231-1963
127 Vol. 6, Issue 1, pp. 114-127
[24]. Emad N. El Qada, Stephen J. Allen, Gavin M. Walker, (2008) “ Adsorption of basic dyes from
aqueous solution onto activated carbons”, Chemical Engineering Journal 135, pp174-184.
[25]. S. L. Pandharipande, Yogesh Moharkar, Raj Thakur, (2012) “Synthesis of adsorbents from waste
materials such as ziziphus jujube seed & mango kernel, International Journal of Engineering
Research and Applications, Vol. 2, Issue4, pp1337-1341.
[26]. S. L. Pandharipande, Umesh Dhomane, Pradip Suryawanshi, Nitin Dorlikar, (2012)
“Comparative studies of adsorbents prepared from agricultural wastes like bagasse, jackfruit peel
& ipomoea fistulos”, A International Journal of Advanced Engineering Research and Studies,
Vol. 1, Issue III, pp214-216.
[27]. Aarti R. Deshmukh, (2013) Minor project report for M. Tech submitted to Rashtrasant Tukdoji
Maharaj Nagpur University, Nagpur.
Authors
S. L. Pandharipande is working as associate professor in Chemical Engineering department
of Laxminarayan Institute of Technology, Rashtrasant Tukadoji Maharaj Nagpur University,
Nagpur. He did his masters in 1985 & joined LIT as a Lecturer. He has co-authored three
books titled ‘Process Calculations’, ‘Principles of Distillation’ & ‘Artificial Neural Network’.
He has two copyrights ‘elite-ANN’ & ‘elite-GA’ to his credit as coworker & has more than 50
papers published in journals of repute.
Aarti R. Deshmukh received the Bachelor of Technology in Chemical Engineering in 2011
from College of Engineering and Technology, Akola, Sant Gadge Baba Amravati University,
Amravati. She is currently pursuing the M. Tech. (Chemical Engineering) from Laxminarayan
Institute of Technology, Rashtrasant Tukadoji Maharaj Nagpur University, Nagpur.