Neural Network Modeling for Ni(II) Removal from Aqueous System Using Shelled Moringa Oleifera Seed Powder as an Agricultural Waste ()
1. Introduction
The pattern of industrial activity alters the natural flow of materials and introduces chemicals in their effluents [1]. Most of these effluents contain toxic substances especially heavy metals. The heavy metals are of special concern because they are non-degradable and thus persistent. The removal of heavy metals from wastewater has recently become the subject of considerable interest due to more strict legislations introduced to control water pollution. Current methodologies such as chemical precipitation, electro floatation, ion-exchange and reverse osmosis have been used for the removal of heavy metals [2]. Activated carbon is also regarded as an effective adsorbent for removal of metal ions from water [3]. However, these processes are economically non-feasible especially for the developing countries [4].
Biomaterials have gained much importance for decontamination of water which involves processes that reduce overall treatment cost through the application of wastes like bagasse pith, wood, saw dust and other agricultural wastes [5-7]. They are particularly attractive as they lessen reliance on expensive water treatment chemicals, negligible requirements of transportation and thus offering genuine, local resources as alternate solutions to tackle local issues of water quality problems. This novel approach is competitive, effective and low cost. Agricultural wastes that are available in large quantities have enough potential to be used as biosorbents in an environment friendly manner. Regeneration of the biosorbents further increases the cost effectiveness of the process thus warrants its future success following the concept of Green Chemistry which is a new principle guiding the next generation products and processes [8].
Ni(II) has been recognized as one of the hazardous heavy metals commonly used in mining, acid battery manufacturing, metal plating etc. [9,10]. Significant quantities of nickel-containing waste water are introduced into water bodies from the effluents of nickel plating plants, silver refineries, zinc based casting industries and storage batteries [11]. Higher concentration of Ni(II) causes cancer of lungs, kidneys, gastrointestinal distress, nausea, vomiting, pulmonary fibrosis, renal edema and skin dermatitis [12]. These harmful effects of Ni(II) necessitate its removal from waste waters before release into streams.
To achieve an optimum management for any control measure, the concept of modeling for an efficient operation and design should be developed. ANN utilizes interconnected mathematical neurons to form a network that can model complex functional relationship [13]. In recent years, ANN have been used as a powerful modeling tool in various processes such as membrane filtration, gas separation, ultra filtration, reverse osmosis etc. [14- 16].
In continuation of our work on biosorption of toxic metals using agricultural waste from waste water [17-20], the present paper describes the abatement of Ni(II) ions from aqueous system using shelled Moringa Oleifera seed (SMOS) powder. Moringa Oleifera, a multidimensional tropical plant that survives in heat, desiccating dryness and destitute soils is deemed to be an efficient biosorbents for metal removal than other biosorbents previously reported. The paper also reports the applicability of a single-layer ANN model using a back propagation (BP) algorithm to predict the removal efficiency of shelled Moringa Oleifera seed powder (SMOS) for Ni(II) ions. Pursuing benchmark comparisons of BP algorithms, a study was conducted to determine the optimization study to determine the optimal network structure. Experimental data were initially distributed to three subsets; training, validation and testing. Finally, output obtained from the ANN modeling was compared with the experimental data. The present piece of work highlights the possibility of the prediction of sorption efficiency for the metal ions from waste water using SMOS in the range of metal concentration with which lab experiments have not been conducted.
2. Materials and Methods
2.1. Biosorbent Preparation
Moringa Oleifera Lam. tree was notified in the nearby area of Dayalbagh Educational Institute and the seeds were collected from the target plant. Seeds were washed thoroughly with double distilled water to remove the adhering dirt, dried at 65℃ for 24 h, crushed and sieved through (105) mesh copper sieves. Shelled Moringa Oleifera seeds (SMOS) were used as biosorbent.
2.2. Biosorption Studies
Sorption studies using standard practices were carried out in batch experiments (triplicate) as a function of biomass dosage (2.0-6.0 g), contact time (10-60 min), volume of the test solution (100-300 mL), metal concentration (10-100 mg/L), particle size (105) and pH (4.5-8.5). A required amount of Ni(II) (Nickel Sulphate, AR grade) was taken in an Erlenmeyer flask and after pH adjustments, a known quantity of dried biosorbent was added and metal bearing suspensions were kept under magnetic stirring until equilibrium conditions were reached. After shaking, the suspension was allowed to settle. The residual biomass sorbed with metal ion was filtered using Whatman 42 filter paper (Whatman International Ltd., Maid stone, England). Filtrate was collected and subjected for metal ion estimation using Flame atomic absorption spectrometer. Percent metal uptake by the sorbent has been computed using the equation: % Sorption = Co – 100Ce/Co, where Co and Ce were the initial and final concentration of metal ions in the solution.
2.3. Statistical Analysis
Batch experiments were conducted in triplicates (N = 3) and data represent the mean values. Regression, correlation coefficients, standard deviations have been calculated using SPSS PC + TM statistical package. For the determination of inter group mean values differences, each parameter was subjected to a student “t” test for determining significance level (p < 0.05).
2.4. Definition of the ANN Model
A neural network is a massively parallel distributed processor made up of simple processing units, which has a natural propensity for storing experimental knowledge and making it available for use. A neuron is an information processing unit that is fundamental to the operation of a neural network. Neural Network Toolbox Neuro Solution 5 ® mathematical software was used to predict the sorption efficiency. Ninety experimental sets were used to develop the ANN model. A single-layer ANN with sigmoid axon transfer function was used for input and output layers. The data gathered from batch experiments were divided into input matrix and desired matrix. The single layer sigmoid network represents functional relationship between inputs and output, provided sigmoid layer has enough neurons. Levenberg-Marquardt algorithm is fastest training algorithm for network of moderate size, therefore, used in the present study.
3. Results and Discussions
3.1. Sorption Studies
Table 1 represents soluble Ni(II) ion concentration after sorption on SMOS powder. Sorption studies led to the standardization of the optimum conditions as: Ni(II) concentration (25 mg/L), contact time (40 min) and volume (200 mL) at pH 6.5 for maximum Ni removal (75.64%).
The results indicate that the SMOS has considerable potential to be used as biosorbent for metal removal from waste water. Studies, therefore, have been planning to predict the efficiency of SMOS for the removal of Ni(II) using ANN model.
Table 1. Soluble Ni(II) ion concentration (µM) after adsorption on SMOS for Ni(II) as a function of metal concentration and biomass dosage at volume (200 mL), pH 6.5 and particle size (105 µM).
3.2. Optimization of the ANN Structure
The prediction of removal efficiency of Ni(II) ions from aqueous system using SMOS are made in the range of metal concentration with which experiments have not been conducted. A training set of ninety experimental data sets was selected to develop the model. ANN model based on single layer recurrent back propagation algorithm for the experimental data was applied to train the neural network. During training, the output vector is computed by a forward pass in which the input is propagated forward through the network to compute the output value of each unit. The output vector is then compared with the desired vector which resulted into error signal for each output unit. In order to minimize the error, appropriate adjustments were made for each of the weights of the network. After several such iterations, the network was trained to give the desired output for a given input vector. The optimum network structure was determined as single layer with 10 hidden neurons (1000 epochs) describing the dynamics of Ni(II) in effluent (Figure 1) respectively.
The sigmoid axon was considered transfer function with 0.7 momentums. The performance of neural network simulation was evaluated in terms of mean squared error (MSE) criterion. The MSE for the training and cross validation data sets were found at the ninth place of decimal. The developed network model was examined for its ability to predict the response of experimental data not forming part of the training program. Figure 2 show the result obtained by the network simulation for both the training and cross validation data sets. The reduction in Ni(II) concentrations were precisely predicted for the training data sets. The development of the proposed ANN model is an effort towards the growing interest in applying ANN modeling technique to the area of biosorption of pollutants from water bodies [21-23].
3.3. Sensitivity Analysis
A sensitivity analysis was conducted to determine the degree of effectiveness of variables. Performance of the groups of input vectors included biomass dosage, Ni(II) ion concentration, contact time and volume of test solution. Series of experiment resulted into the evaluation of the performance based on 50% data for training, 25% data for testing and 25% data for cross validation at 1000 Epoch with 0.70000 momentums. The minimum MSE in the group of four variables determined for training and cross validation were 0.005956571 and 0.00866526823 respectively as shown in the Table 2.
Figure 1. Single layer Neural Network structure for the prediction of the biosorption efficiency.
(a)
(b)
(c)
Figure 2. Graphical representation of MSE value with 1000 Epoch.
The effect of various experimental parameters was studied and compared with performance of ANN model based predictions.
3.4. Effect of Metal Concentration on the Sorption Efficiency
Figure 3 represents the effect of metal concentration on the sorption behavior of Ni(II) on SMOS in the range of metal concentration (10-100 mg/L). Sorption of Ni(II) on SMOS increased with increasing concentration of the metal ion reaching to an optimal level (25 mg/L). Later, an increase in initial concentration decreased the percentage binding. These observations can be explained by