11n13 ijaet0313450 revised

Page 1

International Journal of Advances in Engineering & Technology, Mar. 2013. ŠIJAET ISSN: 2231-1963

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 1

Associate 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

114

Vol. 6, Issue 1, pp. 114-127


International Journal of Advances in Engineering & Technology, Mar. 2013. ŠIJAET ISSN: 2231-1963 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 115

Vol. 6, Issue 1, pp. 114-127


International Journal of Advances in Engineering & Technology, Mar. 2013. ©IJAET ISSN: 2231-1963 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.

116

Vol. 6, Issue 1, pp. 114-127


International Journal of Advances in Engineering & Technology, Mar. 2013. ŠIJAET ISSN: 2231-1963

Figure 1. Neural Network Architecture for model ACM Table 1. Neural network topology for ANN models Number of Neurons Data points RMSE 1st 2nd 3rd Out Training Test Training Test data hidden hidden hidden put data set data data set set layer layer layer layer 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. Model code

Input layer

Table 2. Total data for ANN modeling Initial concentration (mg/ml) Sr. No

1 2 3 4 5 6 7 8 9 10 11 12 13

Ni (II) 1.2345 1.2345 1.2345 0.4938 0.4938 0.8024 0.8024 0.9876 0.9876 0.9876 1.2345 1.2345 0.6173

117

Cr (VI) 0.4418 0.4418 0.4418 0.1944 0.1944 0.3136 0.3136 0.3799 0.3799 0.3799 0.4418 0.4418 0.2209

Type of adsor bent

Amt. of adsor bent (gm)

AMA AMA AMA AMA AMA AMA AMA AMA AMA AMA SCA SCA SCA

1 3 5 3 5 1 5 1 3 5 1 5 1

Equilibrium concentration (mg/ml)

Ni (II) 0.0492 0.2631 0.0754 0.2333 0.2206 0.3081 0.2795 0.0498 0.0962 0.0937 0.2631 0.2714 0.2191

Cr (VI) 0.3342 0.1883 0.2816 0.1 0.0943 0.1815 0.126 0.3537 0.2704 0.2734 0.1883 0.1254 0.0938

Amount of adsorbate adsorbed per unit amount of adsorbent, qe (mg/gm) Ni (II) Cr (VI) 59.26 5.374 16.19 4.224 11.59 1.601 4.342 1.573 2.732 1.001 24.72 6.603 5.229 1.876 46.89 1.309 14.86 1.825 8.939 1.064 48.57 12.67 9.631 3.163 19.9 6.356

Percentage adsorption

Ni (II) 96.02 78.69 93.89 52.75 55.33 61.6 65.17 94.95 90.26 90.52 78.69 78.01 64.49

Cr (VI) 24.33 57.38 36.24 48.56 51.48 42.12 59.81 6.89 28.83 28.02 57.38 71.61 57.54

Vol. 6, Issue 1, pp. 114-127


International Journal of Advances in Engineering & Technology, Mar. 2013. ŠIJAET ISSN: 2231-1963 14 15 16 17 18 19 20 21

0.6173 0.6173 0.8641 0.8641 0.4938 0.8024 1.2345 0.8641

0.2209 0.2209 0.3092 0.3092 0.1944 0.3136 0.4418 0.3092

SCA SCA SCA SCA AMA AMA SCA SCA

3 5 1 3 1 3 3 5

0.2175 0.2148 0.2909 0.2516 0.2313 0.2945 0.292 0.2367

0.0931 0.0917 0.138 0.1112 0.0992 0.1411 0.1551 0.1025

6.662 4.024 28.66 10.21 13.12 8.465 15.71 6.274

2.13 1.292 8.548 3.3 4.759 2.876 4.778 2.067

64.75 65.19 66.33 70.89 53.16 63.3 76.34 72.61

57.85 58.48 55.29 64.04 48.97 55.02 64.89 66.84

Table 3. Actual and Predicted values for the out parameters obtained from ANN model ACM

Equilibrium concentration (mg/ml) Sr. No.

Actual 1st 2nd Ni (II) Cr (VI)

Predicted 1st 2nd Ni (II) Cr (VI)

Amount of adsorbate adsorbed per unit amount of adsorbent, qe (mg/gm) Actual 3rd 4th Ni Cr (II) (VI)

Percentage adsorption

Predicted 3rd 4th Ni Cr (II) (VI)

Actual 5th 6th Ni Cr (II) (VI)

Predicted 5th 6th Ni Cr (II) (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.

118

Vol. 6, Issue 1, pp. 114-127


International Journal of Advances in Engineering & Technology, Mar. 2013. ŠIJAET ISSN: 2231-1963

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

119

Vol. 6, Issue 1, pp. 114-127


International Journal of Advances in Engineering & Technology, Mar. 2013. ŠIJAET ISSN: 2231-1963 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. equlibrium conc. of Ni

0.4

1th Actual output 1th Predicte d output

0.3 0.2

0.1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Data points

Figure 4. Comparison of actual and predicted values of equilibrium concentration of Ni (II) for training data set obtained by model ACM 1th Actual output 1th Predicted output

equlibrium conc. of Ni

0.4 0.3 0.2 0.1 0 Data 2 points 3

1

4

Figure 5. Comparison of actual and predicted values of equilibrium concentration of Ni (II) for test data set obtained by model ACM 0.4

2th Actual output

equlibrium conc. of Cr

0.3 0.2

2th Predicte d output

0.1 0 1 2 3 4 5 6 7 8 9 10 1112 1314 151617 Data points

equlibrium conc. of Cr

Figure 6. Comparison of actual and predicted values of equilibrium concentration of Cr (VI) for training data set obtained by model ACM 0.2

2th Actual output

0.15

2th Predicted output

0.1 0.05 0 1

2 3 Data points

4

Figure 7. Comparison of actual and predicted values of equilibrium concentration of Cr (VI) for test data set obtained by model ACM

120

Vol. 6, Issue 1, pp. 114-127


International Journal of Advances in Engineering & Technology, Mar. 2013. ŠIJAET ISSN: 2231-1963

qe Ni

80

3th Actual output

60 40

3th Predicte d output

20 0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Data points Figure 8. Comparison of actual and predicted values qe of Ni (II) for training data set obtained by model ACM

qe Ni

30

3th Actual output

20

3th Predicted output

10 0 1

2Data points 3

4

Figure 9. Comparison of actual and predicted values qe of Ni (II) for test data set obtained by model ACM 15

4th Predicted output

qe Cr

10

4th Actual output

5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Data points

Figure 10. Comparison of actual and predicted values qe of Cr (VI) for training data set obtained by model ACM 4th Actual output

5

4th Predicted output

qe Cr

10

0 1

2

3 Data points

4

Figure 11. Comparison of actual and predicted values qe of Cr (VI) for test data set obtained by model ACM

% adsorption Ni

150

5th Actual output

100

5th Predicted output

50 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Data points

Figure 12. Comparison of actual and predicted values of % adsorption of Ni (II) for training data set obtained by model ACM

121

Vol. 6, Issue 1, pp. 114-127


% adsorption Ni

International Journal of Advances in Engineering & Technology, Mar. 2013. ©IJAET ISSN: 2231-1963 100

5th Actual output

50

5th Predicted output

0 1

2 3 Data points

4

% adsorption Cr

Figure 13. Comparison of actual and predicted values of % adsorption of Ni (II) for test data set obtained by model ACM 6th Actual output

80 60

6th Predicted output

40 20 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Data points

% adsorption Cr

Figure 14. Comparison of actual and predicted values of % adsorption of Cr (VI) for training data set obtained by model ACM 6th Actual output 6th Predicted output

100 50 0 1

2

3 Data points

4

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 % Relative error Output parameters

Metal ion

Training data point = 17

Test data points = 04

0- ±05 ±05-±10 >±10 0- ±05 ±05-±10 Ni (II) 17 00 00 02 02 Cr (VI) 17 00 00 01 02 Amount of adsorbate adsorbed Ni (II) 16 00 01 00 01 per unit amount of adsorbent (qe) Cr (VI) 14 02 01 00 00 % adsorption Ni (II) 17 00 00 02 01 Cr (VI) 17 00 00 01 03 % Relative error = (Actual value –Predicted value)/ Actual value × 100 Equilibrium concentration

122

>±10 00 01 03 04 01 00

Vol. 6, Issue 1, pp. 114-127


International Journal of Advances in Engineering & Technology, Mar. 2013. ŠIJAET ISSN: 2231-1963 % relative error

1 0 -1

1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17

-2 -3

Data points

Figure 16. % Relative error for equilibrium concentration of Ni (II) for training data set obtained by model ACM

% relative error

0

1

2

3

4

-5

-10 Data points

% relative error

Figurev17. % Relative error for equilibrium concentration of Ni (II) for test data set obtained by model ACM 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

% relative error

Figure 18. % Relative error for equilibrium concentration of Cr (VI) for training data set obtained by model ACM 40 20 0 1

2

3

-20

4

Data points

% relative error

Figure 19. % Relative error for equilibrium concentration of Cr (VI) for test data set obtained by model ACM 10 0 -10

1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17

-20 -30

Data points

Figure 20. % Relative error for qe of Ni (II) for training data set obtained by model ACM

123

Vol. 6, Issue 1, pp. 114-127


International Journal of Advances in Engineering & Technology, Mar. 2013. ŠIJAET ISSN: 2231-1963 % relative error

100 50 0 1

-50

2

3

4

-100 Data points

Figure 21. % Relative error for qe of Ni (II) for test data set obtained by model ACM

% relative error

10 5 0

1

-5

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17

-10 -15

Data points

Figure 22. % Relative error for qe of Cr (VI) for training data set obtained by model ACM

% relative error

50 0 1

-50

2

3

4

-100

-150 Data points Figure 23. % Relative error for qe of Cr (VI) for test data set obtained by model ACM

% relative error

0.4

0.2 0 -0.2

1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17

-0.4 -0.6

Data points

Figure 24. % Relative error for % adsorption of Ni (II) for training data set obtained by model ACM

% relative error

15 10 5 0 -5

1

2 Data points

3

4

Figure 25. % Relative error for % adsorption of Ni (II) for training data set obtained by model ACM

124

Vol. 6, Issue 1, pp. 114-127


International Journal of Advances in Engineering & Technology, Mar. 2013. ŠIJAET ISSN: 2231-1963 % relative error

1 0 -1

1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17

-2 -3

Data points

Figure 26. % Relative error for % adsorption of Cr (VI) for training data set obtained by model ACM

% relative error

10 5 0

-5

1

2

3

4

-10 -15

Data points

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.

125

Vol. 6, Issue 1, pp. 114-127


International Journal of Advances in Engineering & Technology, Mar. 2013. ©IJAET ISSN: 2231-1963 [3]. R. D. Khonde & S. L. Pandharipande, (2011) “Application of Artificial Neural Network for Standardization of Digital Colorimeter”, International Journal of Computer Applications, ICCIA5, 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, pp122131. [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

126

Vol. 6, Issue 1, pp. 114-127


International Journal of Advances in Engineering & Technology, Mar. 2013. ©IJAET ISSN: 2231-1963 [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.

127

Vol. 6, Issue 1, pp. 114-127


Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.