Introduction

Phenols and its derivatives such as chloro and nitro phenols are toxic and carcinogen, usually present in industrial waste water. Very small concentration of phenol may cause vomiting, anorexia, liver and kidney damage, fainting and even mental disturbance. Phenols, long term affecting pollutants can be found in industries which produce chlorophenols that are widely used as fungicides and insecticides in agricultural sector. Phenolic compounds in potable water at the low level of 5 μg/L emit an unpleasant odor and flavor and considered poisonous to aquatic life, plants and human. Ingestion of phenols through water in concentration from 10 to 240 mg L−1 for a long time causes mouth irritation, diarrhea, and excretion of dark urine and vision problems (Hegazy et al. 2013; Asma and Zainal 2009). The WHO (2008) recommended threshold limit of phenol in potable water as 0.001 mg L−1 (Zeng et al. 2009), while US Environmental Protection Agency (EPA) recommended the permissible limit of phenols in the wastewater <1 mg L−1 (Eahart et al. 1977). Phenols can be removed from the aqueous solution through oxidation, precipitation, ion exchange, biodegradation, ultrasonic degradation, solvent extraction, ozonization and decomposition by fenton reagent (Aksu and Yener 2001; Rengaraj et al. 2002). Adsorption is a well-established, low cost and powerful technique for treating domestic and industrial effluents. Recent literature highlighted the extensive use of activated carbon in the field of wastewater treatment (Lotfy et al. 2012; Olafadehan et al. 2012). However, Many other adsorbents like bottom ash, brick-kilm ash, fly ash, peat, soil, rice husk, wood, saw dust, bagasse and carbonized bark are extensively used for removal of organic pollutants (Ademiluy et al. 2009; Aksu and Yener 2001; Hamza et al. 2012; Rengaraj et al. 2002).

For fine tuning and predicting the adsorption mechanism, many software-based models have been used (Carsky and Do 1999; Cuco et al. 2009). In recent years, ANN has become a popular choice among engineers and scientists as one of the powerful tools for predicting contamination and concentration of different effluents and chemicals in drinking water, wastewater and aquifers and energy content in municipal solid waste (Ogwueleka and Ogwueleka 2010). ANN model was used by Bhattacharjya et al. (2007) and Chelani et al. (2004) to characterize the salt water in a coastal aquifer and comparison was made between ANN and multivariate regression.

Orange, as a kind of biological resources, is available in large quantities in many parts of the world (Shan et al. 2012). As the orange residue, orange peel mainly consists of cellulose, hemi-cellulose and lignin in the form of carboxyl and hydroxyl, which results in high affinity to both inorganic and organic moiety. Although many studies in literatures have focused on the modification of orange waste by common chemical modifications such as alkaline and acid treatment, the adsorption capacity and selectivity of heavy metal ions on orange waste are well-documented (Ning-chuan and Xue-yi, 2012; Shan et al. 2012; Li et al. 2007). But removal of organic pollutant using OPA is very limited. In the present study, a simple and economic preparation of the adsorbent from orange peel was performed and adsorption experiments were conducted for removal of phenol from aqueous medium. The equilibrium isotherm data were fitted in Langmuir, Freundlich and DR and Temkin equations. Adsorption kinetics of phenol onto OPA was also studied using kinetic models. Finally the experimental results were analyzed with the help of neural network model.

Materials and methods

Preparation of OPA

Orange peel was collected from fruit juice shop of local market and thoroughly washed with distilled water. Orange peels were dried up in an oven at 80 °C for overnight, cut into small pieces and then carbonized into muffle furnace at 540 ± 2 °C for 1 h. The orange peel ash (OPA) was ground well into a fine powder with a mortar and pestle and sieved through a 250 µm and stored for further use.

Reagents and experimental procedure

A stock solution of phenol was prepared by dissolving 0.5 g of phenol (E. Merck Ltd., India) in double-distilled water in a 500 mL volumetric flask. This was treated as stock solution of phenol with a strength of 1,000 ppm. All the intermediate phenolic solution of lower strength was prepared from this stock solution. About 1.0 g of powder was taken into a 250 cm3 conical flask for batch adsorption process. The pH of the solution was adjusted to the required level, using either HCl (0.1 mol/L) or NaOH (0.1 mol/L) solutions whenever necessary.

Adsorption experiments

Batch adsorption study for different experimental variables (pH; initial concentration; adsorbent dose; stirring rate; contact time and temperature) was carried out by agitating 2.0 g of OPA with 100 mL of synthetic phenol solution in 500 mL conical flask in a temperature-controlled magnetic stirrer (BZMS448 REMI Equipments; Pvt. Ltd., Mumbai, India). At the end of predetermined time interval, the content was filtered and the supernatant was analyzed for residual concentration of phenol spectrophotometrically (APHA et al. 1995) using UV–visible spectrophotometer (Systronics, Vis double beem Spectro 1203).

The amount of phenol adsorbed at equilibrium, qe (mgg−1) was determined using the following equation:

$$q_{\text{e}} = \frac{{(C_{i} - C_{f} )V}}{m}$$
(1)

  • where, V = Volume (L) of the equilibrated solution

  • m = Mass of used OPA (g)

  • C i  = Initial concentration of phenol (mg L−1)

  • C f  = Equilibrium concentration of phenol (mg L−1)

Artificial neural network model

ANN model is a mathematical model made up of simple processing units, which may store experimental data and make it available for further use. The sorption efficiency of OPA was calculated using mathematical software (Neural Network Toolbox Neuro Solution 5 ®). Twenty-four experimental sets were used to develop the ANN model. A multi-layer ANN with sigmoid axon transfer function was used for input and output layers. The data generated from batch experiments and same was used for input and desired matrix. The trained ANN model output results were tested with the experimental output results of phenol adsorption on OPA. The training of the ANN model was done with six neurons in the hidden layer.

Multi-layer perceptron

A typical multi-layer perception is a feed forward ANN model that maps sets onto a set of appropriate outputs. It can learn examples that are “non-linearly separable” like the XOR problem. An additional layer referred to as the ‘hidden layer’ is present between the input and the output layer to handle the non-linearity problem. The number of nodes in the input layer is depended upon the number of variable input parameters in the training dataset. In this present study the same number of nodes was used in the hidden layer as in the input layer for better training.

Back propagation algorithm

The teaching of ANN model was done by back propagation. This algorithm of adjusting the weights of the different network layers starting from the output layer and proceeding downstream is referred to the back propagation algorithm. It was basically worked on the basis of delta rule and required a data set of the desired output for many inputs. Moreover, it is very useful for feed forward network.

Procedure

The net inputs to the network as

$$A(x,w) = \sum\limits_{i = 0}^{n} {x_{i} w_{ji} + \theta }$$
(2)

where \(x_{i}\)= inputs, \(w_{ji}\)= weights of connection, \(\theta\)=threshold = 0.5 (this value was taken because the input data comprises between 0 and 1.)

Then the output was computed by the application of the sigmoid activation function as

$$O_{j} (x,w) = \frac{1}{{1 + e^{{A_{j} (x,w)}} }}$$
(3)

The system error was given by

$$E(x,w,d) = \sum\limits_{j} {\left( {O_{j} \left( {x,w} \right){-}d_{j} } \right)^{2} }$$
(4)

where \(d_{j \, }\)= actual output

Finally the interconnection weights are adjusted using the backpropagation algorithm as

$$\varDelta v_{ik} = - \eta (\frac{\partial E}{{\partial v_{ik} }}) = - \eta (\frac{\partial E}{{\partial x_{i} }} \times \frac{{\partial x_{i} }}{{\partial v_{ik} }})$$
(5)

where

$$\frac{\partial E}{{\partial w_{ji} }} = 2(O_{j} - d_{j} )O_{j} (1 - O_{j} )w_{ji}$$
(6)

and

$$\frac{{\partial x_{i} }}{{\partial v_{ik} }} = x_{i} (1 - x_{i} )v_{ik}$$
(7)

where,\(v_{ik}\) = weights of hidden layer \(\eta\) = momentum, the value of µ was taken to be 0.02

Mean square error (MSE)

The MSE values were calculated by the following equation

$${\text{MSE}} = \frac{1}{N}\mathop \sum \nolimits \mathop \sum \limits_{i = 1} N (T_{i} - A_{i} )$$
(8)

  • where, N = number of data point

  • T i  = Network predicted value at the ith data

  • A i  = Experimental value at the ith data

  • i = an index of the data.

The multi-layer perception model was used for prediction of phenol removal and backpropagation algorithm was used to train the Neural Network (Fig. 1).

Fig. 1
figure 1

Neural network architecture

Full size image

Results and discussion

Adsorbent characteristics

The OPA behaves as neutral at pH zero change. The results are shown in Fig. 2. The physicochemical properties of OPA are summarized in Table 1. To identify the functional groups available on the surface of the investigated adsorbents, the IR spectra were recorded as shown in the Fig. 3a, b. Before adsorption of fluoride OPA showed intense bands at 3,418.60, 2,345.42, 2,363.97, 1,420.20 cm−1 and 1,048.22 cm−1. Among these, the bands at 3,418.60 cm−1 and 1,420.20 cm−1 are attributed to the hydroxyl and amino groups, respectively. But the peak at 3,418.60 cm−1 was shifted to the 3,443.98 cm−1 and the peak at 1,420.20 cm−1 splitted to 1,625 cm−1 and 1,437 cm−1. This is probably due to the interaction between −OH and −NH2 functional groups of adsorbents and phenoxide ions. Characterization of the OPA (before and after adsorption) was also done using SEM micrograph shown in Fig. 4a, b. It is evident from the micrograph that OPA powder was an assemblage of fine particles, which did not have regular fixed shape and size. But after passing phenol solution, adsorbent surface showed uneven surface texture along with lot of irregular surface.

Fig. 2
figure 2

Plot of ∆pH vs. pHi

Full size image
Table 1 Physico-chemical properties of OPA
Full size table
Fig. 3
figure 3

a FTIR spectra of OPA before phenol adsorption. b FTIR spectra of OPA after phenol adsorption

Full size image
Fig. 4
figure 4

a SEM of OPA before phenol adsorption. b SEM of OPA before phenol adsorption

Full size image

Effect of pH

The initial pH of adsorption medium is one of the most important parameters affecting the adsorption process. The adsorption of phenol by OPA was studied at various pH ranging from 3.0 to 7.0 (Fig. 5). It is revealed that the adsorption of phenol by OPA was highest at pH 5.0 and thereafter adsorption was decreased with increasing pH of the medium. It seems to be possible because the surface behaves as positive when pH < pHzpc and adsorption of anion is favored while adsorption of cation is favorable when pH > pHzpc (Gholami et al. 2006). The ionic fraction of phenolate ion (ionsφ) can be calculated from the following equation (Banat et al. 2000):

$$\not\!{c}_{\text{ions}} = 1/1 + 10_{\text{a}}^{\text{PK}} - pH$$
(9)
Fig. 5
figure 5

Effect of pH on adsorption of Phenol using OPA (Initial Phenol concentration of 20.0 mg L−1; adsorbent dose of 2.0 g/100 mL; stirring rate 150 rpm; contact time of 60 min; temperature 25 °C)

Full size image

At low pH, the surface of the OPA is usually protonated and resulted in a stronger attraction for the negatively charged phenolate ions. This is also justified by the pHzpc value (9.1); because maximum adsorption occurs at pH 5 which is just below the pHzpc value. Phenol, being weakly acidic (pKa = 10), partially ionizes in solution and transformed to negatively charged and is directly attracted to the protonated surface of OPA by electro-static force. Unionized phenol molecules would also be attracted, possibly, by physical force. Non-ionized phenol molecules would also be attracted, possibly, by physical force. At high pH, OH− ions would compete with the phenolate ions for sorption sites. Sorption of excess of OH− ions could create a negative charge on the OPA surface resulting repulsion of negatively charged phenoxide ions, and therefore, adsorption is decreased (Uddin et al. 2007). The experimental data and ANN-calculated outputs are compared with ANN model and shows a good performance in the prediction of the experimental data (Fig. 11a).

Effect of adsorbent dosage

To investigate the effect of mass of adsorbent on the adsorption of phenol, a series of adsorption experiments were carried out with different adsorbent dosages at an initial phenol concentration of 50 mg L−1. Fig. 6 shows the effect of adsorbent dosage on the adsorption of phenol. The percentage removal of phenol increased with the increasing absorbent dosage (0.5–4.0 g). It is attributed due to increase in adsorbent surface area and availability of more adsorption sites (Mehrizad et al. 2009; Uddin et al. 2007). The results also revealed that the adsorption efficiency increases with the increasing dose (Rengaraj et al. 2002). The experimental output data are well-fitted with the ANN output data.

Fig. 6
figure 6

Effect of adsorbent dose on adsorption of phenol using OPA (Initial Phenol concentration of 20.0 mg L−1; pH 5.0; stirring rate 150 rpm; contact time of 60 min; temperature 25 °C)

Full size image

Effect of stirring rate

The effect of stirring rate on the adsorption of phenol is shown in Fig. 7. There is a steady increase in the percentage of adsorption with increase in the stirring rate from 100 to 150 rpm. The maximum adsorption occurs at stirring rate of 150 rpm, and thereafter, the adsorption is almost constant. The increase in adsorption at higher stirring probably due to proper contact between the phenolate ions and the adsorbent binding sites and consequently promoting effective transfer of sorbate ions to the sorbent sites. ANN model prediction is found to be matched with the experimental data (Fig. 11d).

Fig. 7
figure 7

Effect of stirring rate on adsorption of phenol using OPA (Initial Phenol concentration of 20.0 mg L−1; pH 5.0; adsorbent dose 2.0 g; contact time 60 min; temperature 25 °C)

Full size image

Effect of initial phenol concentration

The effect of initial concentration on adsorption of phenol is shown in Fig. 8. Maximum adsorption is observed for an initial concentration of 20 mg L−1. Adsorption efficiency shows a decreasing trend with an increasing concentration of phenol from 30 to 60 mg L−1. It is also evident that the adsorption capacity of the sorbent is increased with the increasing phenol concentration while the adsorption yields of phenol showed the opposite trend. It is probably due to increase in mass transfer driving force and therefore the rate at which phenol molecules pass from the bulk solution to the particle surface (Caturla et al. 1998; Imagawa et al. 2000). The experimental data and ANN-calculated output is found to be nicely matched (Fig. 11b).

Fig. 8
figure 8

Effect of Initial concentration on adsorption of phenol using OPA. (pH—5.0; adsorbent dose 2.0 g; stirring rate 150 rpm; contact time 60 min; temperature 25 °C)

Full size image

Effect of contact time

The adsorption data for the uptake of phenol versus contact time is presented in Fig. 9. This result indicates that up to 90–95 % of the total phenol uptake occurs in the first rapid phase (60 min) and thereafter the adsorption rate is decreased. The higher adsorption rate at the initial period (first 40 min) may be due to large number of vacant sites on the adsorbent and as a result there exist increased concentration gradients between adsorbate in solution and adsorbate on adsorbent surface (Uddin et al. 2007). ANN model prediction in accordance with the experimental data (Fig. 11e).

Fig. 9
figure 9

Effect of contact time on adsorption of phenol using OPA (Initial concentration 20.0 mg L−1; pH 5.0; adsorbent dose 2.0 g; stirring rate 150 rpm; temperature 25 °C)

Full size image

Effect of temperature

Batch adsorption experiments were carried out at different temperatures ranging from 298 to 313 K (Fig. 10). It is found that with an increase in temperature from 298 to 313 K, the adsorption capacity of phenol onto OPA is decreased from 0.973 to 0.773 mgg−1. The decrease of adsorption capacity at higher temperature indicates that lower temperature favors the phenol adsorption onto OPA and the adsorption is an exothermic process. At higher temperature entropy of the adsorbed molecule is probably increased and subsequently escapes from the solid adsorbent surface to the bulk phase of solution (Bhatti et al. 2010). The experimental data are fitted well with the ANN model (Fig. 11f; Table 2).

Fig. 10
figure 10

Effect of temperature (K) on adsorption of phenol using OPA (Initial concentration 20.0mgL−1; pH—5.0; adsorbent dose 2.0 g; Stirring rate 150 rpm; contact time 60 min.)

Full size image
Fig. 11
figure 11

Agreement between ANN outputs and experimental data as a function of a pH, b initial concentration (mg L−1), c adsorbent dose, d stirring rate, e contact time (min) and f temperature (K)

Full size image
Table 2 Equations of Isotherm models used in adsorption of phenol onto OPA
Full size table

Adsorption isotherm

Equilibrium study provides information on the efficiency of the adsorbent. An adsorption isotherm is usually characterized by certain constant values, which express the surface properties and affinity of the adsorbent. The most widely used isotherm equations for modeling of the adsorption of phenol are as follows:

$$\frac{1}{{q_{\text{e}} }} = \frac{1}{{K_{\text{L}} q_{\text{m}} }}\frac{1}{{C_{\text{e}} }} + \frac{1}{{q_{\text{m}} }}$$
(10)
$$\log q_{\text{e}} = \log K_{\text{f}} + \frac{1}{n}\log C_{\text{e}}$$
(11)
$$q_{\text{e}} = q_{\text{DR}} \exp ( - K_{\text{DR}}^{{\varepsilon^{2} }} )$$
(12)
$$q_{\text{e }} = B\ln K_{\text{T}} + B\ln C_{\text{e}}$$
(13)

where qmax (mgg−1) and KL (L mg−1) are Langmuir parameters related to maximum adsorption capacity and free energy of adsorption, respectively. KF and n are the Freundlich constants that indicate adsorption capacity and adsorption intensity, respectively. RT/bT = B, where T is the temperature (K), and R is the ideal gas constant (8.315 Jmol−1K−1) and KT (L min−1) and bT are constants. qm (mgg−1) is related to maximum adsorption capacity and β is a constant related to the mean free energy of adsorption per mole of the adsorbate (mol2J−2), and ε is the polanyi potential. The constant, β gives an idea about the mean free energy Es (kJ mol−1) of adsorption per molecule of the adsorbate when it is transferred to the surface of the solid from infinity in the solution. Ce is the equilibrium concentration in the aqueous solution and qe is the equilibrium adsorption capacity of adsorbent. The essential features of the isotherm can be expressed in terms of a dimensionless constant separation factor (RL) that can be defined by the following relationship (Anirudhan and Radhakrishnan 2008):

$$R_{\text{L}} = \frac{1}{{1 + K_{\text{L}} C_{0} }}$$
(14)

According to both Table 3 and Fig. 12, the Langmuir isotherm model shows excellent fitness to the experimental data with high correlation coefficient at all temperatures. The maximum phenol adsorption capacity onto OPA is found to be 3.55 mgg−1 at 303 K.

Table 3 Adsorption isotherm constants for adsorption of phenol onto OPA
Full size table
Fig. 12
figure 12

Comparison between the measured and modeled isotherm profiles for the adsorption of phenol by orange peel ash (experimental conditions: Initial concentration 20.0 mg L−1; pH—5.0; adsorbent dose 2.0 g; stirring rate 150 rpm; contact time 60 min, temperature 298 K)

Full size image

The fitness of equilibrium data (Fig. 12) in Langmuir isotherm advocates the monolayer coverage of phenol onto OPA (Uddin et al. 2007). The essential features of Langmuir isotherm can be expressed in terms of a dimensionless constant separation factor (RL). RL value indicates the favorable adsorption of phenol onto OPA (Maheswari et al. 2009). Again from Table 4, it is clear that OPA is a good adsorbent for phenol among the other reported adsorbents. The Freundlich constant, KF indicates the adsorption capacity of OPA and the value of KF is 1.011 mgg−1. Furthermore, the value of ‘n’ at equilibrium is 1.90, indicating favorable adsorption (Slejko 1985). From D–R isotherm the value of the adsorption energy is found to be 1.83 kJ mol−1 indicating the physisorption mechanism. Again from Temkin constants (B) which is related to the heat of adsorption is moderated. However, other equilibrium constant (KT) data (5.69 L/mg) suggest the maximum binding energy between OPA and phenol molecules (Table 5).

Table 4 Maximum adsorption capacities, qmax (mg/g), for the adsorption of phenolic compounds by various adsorbents
Full size table
Table 5 Kinetic model constants for adsorption of phenol onto OPA
Full size table

Adsorption kinetic models

Kinetic models are used to evaluate the rate of the adsorption process and rate-controlling step. In the present communication, the kinetic data obtained from batch studies are analyzed using the following kinetic models (Table 6).

$${\text{Pseudo first}} - {\text{order}}:\;\;\log (q_{\text{e}} - \;q_{\text{t}} ) = \;\log q_{\text{e}} \; - \;\frac{{k_{1} t}}{2.303}$$
(15)
$${\text{Pseudo second}} - {\text{order}}:\;\;\frac{t}{{q_{\text{t}} }} = \frac{1}{{k_{2} q_{\text{e}}^{2} }} + \frac{1}{{q_{\text{e}} }}t$$
(16)
Table 6 Equations of kinetic models used in adsorption of phenol onto OPA
Full size table

The intra-particle diffusion:

$$q_{\text{e}} = k_{\text{id}} t^{0.5} + C$$
(17)

where qe and qt are concentrations of phenol at equilibrium and at time t. k1 and k2 are constants of the pseudo-first- and pseudo-second-order kinetic model. Kid is the constant of intra-particle diffusion model and C is related to boundary layer effect. The results of three kinetic models are shown in the Table 5 to find out the best fit rate of reaction for the adsorption of phenol onto OPA. According to linear regression correlation coefficients (Table 5), the rate of adsorption is found to follow the pseudo-second-order kinetic model and the theoretical qe value is closer to the experimental qe value. In the view of the present results, it can be said that the pseudo-second-order kinetic model provides a good correlation for the adsorption of phenol onto OPA compared to that of pseudo-first-order model (Figs. 13, 14). Therefore, Pseudo-second-order model is highly applicable for this adsorption process. To identify the diffusion mechanism, the kinetic results are analyzed using the intra-particle diffusion model (Weber and Morris 1963).

Fig. 13
figure 13

Pseudo-first-order adsorption kinetic model for phenol adsorption with OPA

Full size image
Fig. 14
figure 14

Pseudo-second-order adsorption kinetic model for phenol adsorption with OPA

Full size image

Thermodynamic parameters

The Gibbs free energy (ΔG°) for the adsorption of phenol onto OPA is calculated following the equation (ΔG° = −RT ln K) and data are presented in Table 7. The values for ΔH° and ΔS° are determined from the slope and intercept of the plot of ΔG° vs. T (figure not shown) and are also listed in Table 7. The negative value of ΔG° at all temperature indicates the feasibility as well as spontaneity of the phenol adsorption on OPA. Decrease in the value of ΔG° with increase in temperature suggests that lower temperature favors the adsorption (Saha et al. 2010). The negative value of ΔH° implies that the adsorption phenomenon is exothermic while the negative value of ΔS° suggests the process is enthalpy oriented (Saha et al. 2010).

Table 7 Thermodynamic parameters for phenol adsorption onto OPA
Full size table

Activation energy

The activation energy (Ea) usually provides important information on the mechanism of adsorption reaction. Using the pseudo-second-order rate constant, k2 the activation energy, Ea for the adsorption of phenol on OPA is determined using the Arrhenius equation (18).

$$\ln k = \ln A - \frac{{E_{\text{a}} }}{\text{RT}}$$
(18)

where k is the rate constant, A is the Arrhenius constant, Ea is the activation energy (kJ mol−1), R is the gas constant (8.314 J mol−1K−1) and T is the temperature (K). By plotting lnk2 vs. 1/T, Ea is obtained from the slope of the linear plot (Fig. 15). The activation energy is very significant to evaluate whether the entire adsorption reaction is physisorption or chemisorptions. If the value for activation energy lies between 8 and 16 kJ mol−1, it is chemisorptions, and when it is below 8 kJ mol −1, it is physisorption (Hai-jun et al. 2009). Here the Ea value is −18.001 kJ mol−1. The measured Ea value suggests that the adsorption may be a physical adsorption.

Fig. 15
figure 15

Plot of lnk2 vs. 1/T

Full size image

Tested with ANN model

The trained ANN model is tested and validated with the experimental results to estimate the phenol concentration. The network is trained with given input data set (Table 8). The training phase is completed after 6,00,000 epochs (Fig. 16). The lower value of MSE indicates the degrees of error that means network gives correct output. It is also clear from the Fig. 16 that with the increasing epoch the network is trained more accurately and subsequently more accurate and perfect output is achieved.

Table 8 The outputs of MLP
Full size table
Fig. 16
figure 16

Relation between the MSE and number of epochs

Full size image

After the training phase the network shows optimum result describing the dynamics of phenol adsorption from aqueous solution. Finally in testing phase, the results show that the network output is matched with the experimental output (Fig. 11a–f). It is observed from Fig. 16 that MSE value is 0.0017 for one epoch that decreases with the increasing number of epochs and found minimum (0.0006) at 11 numbers of epochs. So the 11 number of hidden layer may be considered as optimum for this ANN model.

Conclusion

It has been found that the OPA has enough potentiality to remove phenol from water. The operational parameters such as pH, initial phenol concentration, adsorbent dose, contact time, stirring rate and temperature have significant influence on the adsorption efficiency of OPA. The maximum adsorption capacity is 3.55 mgg−1 at 20 mg L−1 initial phenol concentration. The multi-layer ANN modeling technique can be applied to optimize the process. The Back Propagation Algorithm (BPA) is found to be the best algorithm with a minimum mean squared error (MSE) for training 0.00528. The temperature has strong influence on the adsorption process and the maximum removal is observed at 298 K. The kinetic study reveals that the adsorption process usually follows pseudo-second order. Langmuir isotherm model is in accordance with the experimental data. Moreover, thermodynamic parameters are in favor of exothermic and spontaneous nature of adsorption of phenol onto OPA.