Modeling and evaluation of electro-oxidation of dye wastewater using artificial neural networks

Abstract

Treatment of CBSOL LE red wool dye containing wastewater by an electro-oxidation (EO) method was investigated using Ti/RuO₂ electrode. The performance of the EO system was evaluated in terms of % dye degradation (Y₁) and % colour removal (Y₂) along with important operating cost parameters such as energy consumed (Y₃) at three EO process parameters: pH, current (i) and time (t). ANNs were applied for the modeling of the EO process, and optimization was performed by using multi response optimization by desirability function approach of Central composite Design (CCD) with stimulated data obtained from ANNs. Modeling for the treatment of CBSOL LE red wool dye wastewater by the EO process was done successfully by ANNs, and optimization by CCD vividly underscores interactions between variables and their effects for the degradation of CBSOL LE red wool dye by the EO process. At the optimum conditions, the actual % dye degradation (Y₁), % color removal (Y₂) and energy consumed (Y₃) were 89.87%, 96.71%, 2.029 Wh respectively. The predictions agree well with the experimental results. It was found that both the mechanisms of EO treatment i.e. direct oxidation and indirect oxidation are responsible for the dye degradation/color removal. Color was found to be nearly completely removed, whereas 10.13% of dye is present in the treated wastewater.

---

1. Introduction

A substantial amount of water is consumed during dyeing/printing and finishing operations of textile processing.^1,2 Consequently, a large amount of high strength colored wastewater is produced.³ Various types of dyes such as basic, acid, dispersed, direct, reactive, azoic and sulphurous dyes are usually applied during the textile processing according to the method and fixation-type adopted for the textile fiber.

CBSOL LE dyes are reactive dyes developed for wool dyeing.⁴ These dyes form covalent bond with wools amino acids during the dyeing process. Also, reactive dyes have been reported to be recalcitrant, and can be carcinogenic to human.¹ Therefore, these effluents are toxic to the environment and must be treated before discharge to water bodies.

Generally, physico-chemical treatment methods like chemical coagulation, adsorption processes and membrane filtration^5,6 are not preferred for the treatment of such type of dye containing effluents due to expensive chemical coagulants, adsorbents, membranes cleaning, and production of large volume of secondary pollutants. Whereas, non-biodegradability of reactive dyes and high energy requirement, restricts the use of biological methods for the treatment of dye containing wastewater.^7,8 These disadvantages i.e. large volume of secondary pollutants generation and non-biodegradability of dye bearing wastewater can be overcome by the use of electro-oxidation (EO) process. EO does not require adding large amount of chemicals to wastewater, as for the case of chemical oxidation, with no generation of secondary pollutants.⁹

Various authors have reported the EO treatment of textile wastewater^10–28 with good removal efficiencies using varieties of anodes like aluminum, iron, graphite, lead/lead dioxide, nickel, platinum and TiO₂/RuO₂ coated Ti. However, anode instability and additional treatment required for the removal of lead and other anodic material from the treated wastewater were the major drawback with the aluminum, iron, graphite, lead/lead dioxide, nickel and platinum anodes. Whereas, TiO₂/RuO₂ coated Ti anodes are dimensionally stable, and no subsequent treatment is required for treated effluent. Kim et al.¹² and Carneiro et al.²⁶ studied the electrochemical treatment of dye wastewater using Ti/RUO₂ anode. However, the detailed dye mineralization mechanism is lacking. Recently, Ramírez et al.²⁷ and Diagne et al.²⁸ reported dye wastewater treatment using a newly invented boron doped diamond (BDD) anode. BDD anodes reported to have high overvoltage and oxidation potential than Ti/RUO₂. Since, BDD anodes are very expensive than Ti/RUO₂, therefore, more study is needed to explore the mineralization mechanism and treated effluent characteristics with Ti/RUO₂ anode.

Furthermore, the studies reported in the open literature for dye wastewater treatment by EO method elucidates the treatment efficiency in terms of color removal/dye removal and/or COD removal. Generally, color removal is incorrectly interpreted with dye degradation/removal. It is well known that change in the pH of dye solution may alter the color intensity.²⁹ Therefore, interpretation of dye removal/degradation with the color removal is not always true. The dye may present in wastewater in its original/degraded structure depending on treatment method applied. Hence, observation of both color removal and dye degradation simultaneously explains the true treatment efficiency. Moreover, During the EO process, dye is first degraded to some intermediates, and these intermediates are further oxidized to carbon dioxide and water³⁰ by varieties of electrochemically generated oxidants such as Cl₂, HOCl and ClO⁻, ˙OH radicals and ˙OCl radicals etc. (discussed in detail in Section 2). During the oxidation of dye mediated by chlorine species chlorinated organic compounds are formed, which are toxic and may be present in the treated dye wastewater. Whereas, ˙OH radicals has been reported to be having high standard redox potential (E° (OH/H₂O) = 2.80 V/SHE), and in oxidation of dye mediated by ˙OH radicals, no such toxic compounds are formed.

In the present study, EO of CBSOL LE red wool dye containing wastewater was studied using RuO₂ coated Ti electrode (Ti/RuO₂), not previously reported. Effects of EO processes parameters like pH, current (i) and time (t) on % dye degradation (Y₁), % color removal (Y₂) along with one of the important operating cost parameter such as energy consumed (Y₃) were investigated. Artificial neural networks (ANNs) was used for modeling of EO process, and multi response optimization using desirability function approach of central composite design (CCD) was used for the optimization of EO of dye wastewater in terms of maximization of response Y₁ and Y₂ and simultaneous minimization of response Y₃. The modeling of such system having multi variables (pH, i and t in present case) and multi response (Y₁, Y₂ and Y₃ in present case) is quite complex, and obviously cannot be solved by simple linear multivariate correlation.²⁸ Since, Modeling based on ANNs does not require the mathematical description of the phenomena involved in the process, therefore, ANNs was used in the present study for modeling of EO process. Furthermore, Mechanism of oxidation i.e. by direct oxidation and/or mediated by chlorine species and/or ˙OH radicals or by all was explored.

2. Theoretical background

In EO process, organics are oxidised by direct and/or indirect oxidation.⁹ Hydroxyl radicals (˙OH) are generated at anode surface and in the bulk of solution. In direct anodic oxidation, the generated ˙OH at anode are adsorbed in the oxide lattice of anode, and organics are oxidised at anode due to these adsorbed ˙OH. The adsorption of ˙OH in the oxide lattice of anode is favoured in acidic media.^9,31 Furthermore, these generated ˙OH are highly unstable and are available for very short time. These ˙OH, if not adsorbed, are rapidly transformed to H₂O₂ and HO^˙₂ (eqn (4) and (5)).³² The oxidation potential of ˙OH is very high in comparison to H₂O₂ and HO^˙₂.

The indirect oxidation of organics is mediated by electrochemically generated oxidants during the electrolysis. In the presence of sodium chloride, various chloro-oxidant species (Cl₂, HOCl and ClO⁻) are generated at anode.^9,30,33,34 The types of chloro-oxidant species formed depend on pH of solution, whereas, stability depends on chloride ion concentration, ionic strength and the temperature. Cl₂ is found at highly acidic pH (pH < 3.0), at 3 < pH < 5, HOCl is found to dominate, whereas, OCl⁻ is the predominant species for pH ≥ 7 and reported having highest oxidation capability among all chlorine species.^31,33 At alkaline conditions, the ClO₃⁻ and ClO₄⁻ ions are also generated.³⁵ Also, hydrogen peroxide, ozone and HO^˙₂ are produce through following reaction (eqn (1)–(5)).^29,32

O₂ + 2H⁺ + 2e⁻ → H₂O₂ (1)

2H₂O + 2O₂ → 2O₃ + 4H⁺ + 2e⁻ (2)

2H₂O → 2˙OH + 2H⁺ + 2e⁻ (3)

2˙OH → H₂O₂ (4)

˙OH + H₂O₂ → HO^˙₂ + H₂O (5)

In addition, oxygen evolution can compete with ˙OH radicals generation at the anode via the following reaction:

2H₂O → O₂ + 4H⁺ + 4e⁻ (6)

Except these mediators, nascent oxygen, free chlorine and ˙OCl free radicals are generated in a typical set-up of electrolytic oxidation and mediate the oxidation of organics.^20,29,30

3. Materials and methods

3.1. Dye wastewater and experimental setup

Dye wastewater was prepared by dissolving 0.1 g of CBSOL LE red wool dye in one litre (L) of distilled water. CBSOL LE red wool dye was supplied from the Grasim Industries Ltd., Nagda, Madhya Pradesh, India. The dye wastewater was prepared freshly whenever it required.

To conduct the EO experiments for treatment of CBSOL LE red wool dye wastewater, a cuboid shape batch EO reactor of 1.5 L working volume was fabricated with acrylic plexi glass sheet of thickness 5 mm. Ti/RuO₂ electrodes purchased from Titanium Tantalum Products Limited, Chennai, India, having the dimension of 100 mm × 85 mm × 1.5 mm was used as anode, whereas aluminium (Al) plate of same dimensions was used as cathode. Two pairs of such electrodes connected in parallel mode with 1 mm inter-electrode spacing were used in the present study.

During the experiments, current was maintained constant using a precision digital direct current power supply (DIGITECH, Roorkee, India, Model: 4818A10; 0–20 V, 0–5 A). Magnetic stirrer was used to agitate the dye wastewater sample in the reactor.

3.2. Experimental procedure and analysis

Before start of an experiment, the dye wastewater pH was adjusted with 0.1 N NaOH/HCl solutions as per the condition of particular run, and requisite amount of NaCl was added to facilitate the indirect oxidation and simultaneously to improve the electrical conductivity of wastewater sample. Power supply was switched on, and electrolysis time (t) was measured from this moment. Current (i) was maintained constant during the EO experiment. After the desired t (as per the experiment) power supply was switched off and samples were drawn from the reactor.

Dye contains a chromogen–chromophore with auxochrome. Chromogen is the aromatic structure, while chromophore is a color giver. The auxochrome are bonding affinity groups. Scanning of CBSOL LE red wool dye solution with UV visible spectrophotometer shows two peaks at different λ_max values of 504 and 270 nm. Peak at λ_max = 504 nm is due to red color produced, while peak at λ_max = 270 nm is due to the aromatic ring associated with dye.²⁹ Hence, it can be interpreted that presence of peak measured at λ_max = 270 nm shows occurrence of respective amount of dye/degraded dye in the dye solution. Therefore, samples of CBSOL LE red wool dye before and after EO treatment were analyzed with double beam UV visible spectrophotometer (HACH, DR 5000, USA) at λ_max value of 270 nm and 504 nm to find of % dye degradation (Y₁) and % color removal (Y₂), respectively. Energy consumption (Y₃) during the EO treatment was calculated by the following equation:

Y₃ = i × V × t (7)

where, i = current; V = voltage and t = treatment time.

3.3. Modeling and optimization

Data from the experiments were used for the ANN modeling. ANN's are simulated by biological neural systems. Neurons are simple processing units of ANN. An artificial neuron is a computational model inspired from natural neurons. Each network of ANN consists of artificial neurons which are grouped in a layer and have parallel relation to each other. Two types of ANNs commonly used are Kohonen self-organizing mapping and multilayered feed forward neural network trained by back-propagation algorithm.³⁶ The majority of ANN architectures are feed-forward networks which are mostly trained from the input data using error back-propagation algorithm. ANN which using the back-propagation algorithm for learning the appropriate weights is one of the most common models used in ANN.³⁷ The most widely used transfer function for the input and hidden layers are the sigmoid transfer function.^38,39

All ANN calculations carried out using Matlab 7.6 mathematical software with ANN toolbox. In the present study, two-layered network with the first layer was tan-sigmoid and the output layer transfer function is linear with backpropagation. Fig. 1 shows the ANN architecture with input, output and hidden layers for the present study. The number of input and output neurons effectively represents the number of variables used in the prediction and the number of variables to be predicted, respectively. The hidden layers act as feature detectors and, in theory, there can be more than one hidden layer. The universal approximation theory, however, suggests that a network with a single hidden layer with a sufficiently large number of neurons can interpret any input output structure.^40–42

Fig. 1 ANN architecture with input and output layers.

Central composite design (CCD) was used for optimization. The stimulated data obtained by ANN model, was analyzed using Design Expert trial version.

The outputs (responses) Y_i which are the functions of input factors X₁, X₂,……X_i,……X_f, are obtained from the following relationship:

Y_i = Φ(X₁, X₂, X₃,…………,X_i,…………X_f) (8)

The above relation between responses and the input factors are considered as quadratic response model. The relevant model terms are identified using non-linear regression analysis to fit the responses according to simulated results and input factors. The model being used is best fitted in second-order polynomial eqn (9).

where, Y is response; b₀, b_i, b_ii, b_ij are constant coefficients and X_i the uncoded independent variables.

For the present case, a multi response optimization is used due to involvement of three responses (% dye degradation Y₁, % color removal Y₂ and energy consumed, Y₃) and three input process parameters, namely pH, current (i) and time (t).

Five level factor coding were followed, and coded as −2 (low) and +2 (high). The design level limits used in the present work are given in Table 1.

Table 1 Range of variables and coded levels

Variables	Range of actual and coded variables
	−2	−1	0	+1	+2
pH	4	5.50	7	8.50	10
Time, t (min)	10	30	50	70	90
i (A)	0.25	0.50	0.75	1.00	1.25

---

4. Results and discussion

4.1. ANN modeling

In the present work, pH, time and current were an input variables and % color removal, % dye degradation and energy consumption were output variables. Total 117 experiments were conducted and used to feed the ANN structure. The data sets were divided into training, validation and test subsets, each of which contains 70%, 15% and 15% samples, respectively. The validation and test sets, for the evaluation of the validation and modeling power of the nets, were randomly selected from the experimental data. Selection of topology i.e. optimum number of hidden layer neurons in the ANN architecture, number of hidden layers in the ANN architecture and the nature of the transfer function are most important steps.⁴³ Preprocessing of the inputs and output was done by randomization to make the neural network training more efficient. Hyperbolic tangent ‘TANSIG’ being a sigmoid transfer function was chosen for the input to hidden layer mapping while a purely linear transfer function ‘PURELIN’ was chosen for the hidden layer to the output layer mapping.

To obtain optimum number of neurons, a series of topologies was used, in which number of neurons was varied from 2 to 10. Mean square error was an error function in these topologies. Backpropagation algorithm minimized the mean square error between the observed and the predicted values. Number of neurons of the hidden layer was adjusted with minimum mean square error.³⁸ Fig. 2 indicated that the optimized neurons for the process were eight.

Fig. 2 Effect of the number of hidden layer neurons on the performance of neural network.

The regression plots of the trained network were shown in Fig. 3. The training versus target gives regression coefficient of 0.995, along with training, validation and test of all data sets regression coefficient value 0.996, 0.992, 0.995 respectively. Which implied that training of the ANN model was done accurately and model was ready to simulate the outputs from a given inputs.

Fig. 3 The regression plots of training versus target.

4.2. Statistical analysis with CCD

For the regression analysis of simulated data, the statistical Design-Expert software version 6.06 (STAT-EASE Inc., Minneapolis, US) was used. The results of the (% dye degradation; Y₁, % color removal; Y₂ and energy consumed; Y₃) for the electro-oxidation of CBSOL LE red wool dye were measured according to the design matrix and are listed in Table 2.

Table 2 Full factorial CCD matrix used, and stimulated data from ANNs of responses

Std	pH	t (min)	i (A)	% dye removal, Y1	% color removal, Y2	Energy consumed (Wh), Y3
1	5.50	30.00	0.50	41.55	59.03	0.995
16	7.00	50.00	0.75	62.24	79.81	2.297
14	7.00	50.00	1.25	80.49	94.32	7.283
3	5.50	70.00	0.50	62.89	81.45	2.221
19	7.00	50.00	0.75	62.24	79.81	2.297
10	10.00	50.00	0.75	68.02	89.12	3.124
8	8.50	70.00	1.00	97.77	98.98	4.825
9	4.00	50.00	0.75	80.63	90.84	2.844
6	8.50	30.00	1.00	93.68	92.04	3.31
11	7.00	10.00	0.75	43.79	50.85	0.602
18	7.00	50.00	0.75	62.24	79.81	2.988
13	7.00	50.00	0.25	30.82	53.46	0.813
15	7.00	50.00	0.75	62.24	79.81	2.988
4	8.50	70.00	0.50	52.79	75.58	2.546
12	7.00	90.00	0.75	80.49	92.53	5.377
20	7.00	50.00	0.75	62.24	79.81	2.988
17	7.00	50.00	0.75	62.24	79.81	2.988
7	5.50	70.00	1.00	89.96	98.25	6.217
5	5.50	30.00	1.00	83.72	76.41	3.005
2	8.50	30.00	0.50	32.61	55.33	1.357

---

The sequential F-test and other adequacy measures were exploited for selecting the best model.⁴⁴

A manual regression process was used to fit the second order polynomial (eqn (9)) to the simulated data and to identify the relevant model terms. To come to a decision about the adequacy of the model for the responses Y₁, Y₂ and Y₃, the sequential model sum of squares and model summary statistics were carried out. p values for all the responses (Y₁, Y₂ and Y₃) were found less than 0.01 and the quadratic model was suggested. This means that at least one of the terms in the regression equation had a significant correlation with the output variable.⁴⁵

The residuals are in the proximity of the straight diagonal line between the actual and predicted values for all the three responses. Therefore developed models are considered to be adequate because the residuals for the prediction of each response are minimum.

Table 3 shows the regression analysis obtained from the design expert software provides ANOVA for the % dye degradation Y₁, % color removal Y₂ and energy consumed, Y₃ for the electro-oxidation treatment of CBSOL red wool dye, with a model F-value of 120.19, 91.65, 47.58 for Y₁, Y₂ and Y₃ respectively, implying that the model is significant.

Table 3 Anova suggested by CCD for the % dye degradation, % color removal and energy consumption

Source	% dye degradation					% color removal					Energy consumption
	Sum of squares	DF	Mean square	F-value	Prob > F	Sum of squares	DF	Mean square	F-value	Prob > F	Sum of squares	DF	Mean square	F-value	Prob > F
Model	7.2 × 10^−4	9	8.0 × 10^−5	120.19	<0.0001	3981.54	9	442.39	91.65	<0.0001	54.87	9	6.1	31.47	<0.0001
pH	9.1 × 10^−6	1	9.1 × 10^−6	13.58	0.0042	0.70	1	0.70	0.15	0.7110	0.0016	1	0.0016	0.008	0.9294
t	1.1 × 10^−4	1	1.1 × 10^−4	164.00	<0.0001	1497.88	1	1497.88	310.31	<0.0001	17.41	1	17.41	89.89	<0.0001
i	4.5 × 10^−4	1	4.5 × 10^−4	682.52	<0.0001	1936.22	1	1936.22	401.12	<0.0001	33.57	1	33.58	173.32	<0.0001
pH^2	1.2 × 10^−5	1	1.2 × 10^−5	18.07	0.0017	171.86	1	171.86	35.60	0.0001	0.049	1	0.05	0.25	0.6248
t^2	2.6 × 10^−6	1	2.6 × 10^−6	3.94	0.0752	96.40	1	96.40	19.97	0.0012	0.05	1	0.052	0.27	0.6143
i^2	5.8 × 10^−5	1	5.8 × 10^−5	87.12	<0.0001	49.85	1	49.85	10.33	0.0093	2.42	1	2.42	12.49	0.0054
pH × t	1.2 × 10^−6	1	1.2 × 10^−6	1.87	0.2010	36.42	1	36.42	7.55	0.0206	0.37	1	0.37	1.94	0.1938
pH × i	1.7 × 10^−5	1	1.7 × 10^−5	25.90	0.0005	84.05	1	84.05	17.41	0.0019	0.39	1	0.39	2.03	0.1846
t × i	4.3 × 10^−5	1	4.3 × 10^−5	64.46	<0.0001	24.12	1	24.12	5.00	0.0494	0.66	1	0.67	3.44	0.0929
Residual	6.7 × 10^−6	10	6.7 × 10^−7			48.27	10	4.83			1.93	10	0.19
Lack of fit	6.7 × 10^−6	5	1.3 × 10^−6			48.27	5	9.65			1.30	5	0.26	2.04	0.2259
Pure error	0.000	5	0.000			0.000	5	0.000			0.63	5	0.13
Cor total	7.3 × 10^−4	19				4029.81	19				56.81	19

---

The model summary statistic showed that the value of regression coefficient R² 0.9908, 0.9880 and 0.977 for Y₁, Y₂ and Y₃, respectively. This advocates a good correlation between the observed and predicted values.

For the % dye degradation Y₁, % color removal Y₂ and energy consumed, Y₃ the adequate precision were 38.68, 30.43 and 20.53 respectively. Adequate precision expresses the signal to noise ratio, and adequate precision ratio above 4 indicates adequate model competence i.e. model is efficient in navigating the design space.⁴⁶

4.2.1. Effects of parameters and removal mechanism. To explore the effects of various parameters of EO i.e. pH, time (t) and current (i) on responses Y₁, Y₂ and Y₃ for EO of CBSOL LE red wool dye, 3-D response surface graph obtained from RSM were studied.

For pH value from 4 to ≈8.0, it can be observed in Fig. 4a that % dye degradation, Y₁ increases with increase in i value from 0.25 to ≈0.75 A. Further increase in i value beyond ≈0.75 showed marginal effect on the % dye degradation. While, at each pH beyond 8.0 increasing i value first increases Y₁ and then Y₁ becomes constant showing nearly complete removal of dye. It can also be observed that increasing pH at high value of i (i > 0.75) affected Y₁ marginally, but at lower i value (i < 0.75) increasing pH significantly lowered the Y₁ upto pH value ≈8.0, and for each pH > 8.0 increased Y₁ was observed. Therefore, it can be concluded that increasing i value at highly acidic and basic pH leads to increase in % dye degradation, Y₁, and at highly basic pH complete dye removal is observed at i = 0.75 A.

Fig. 4 Three dimensional response surface graph for the electro-oxidation of CBSOL LE red wool dye wastewater (a) % dye degradation versus i and pH (b) % color removal versus i and pH (c) % dye degradation versus pH and t (d) % color removal versus pH and t.

Same trend of removal was observed for % color removal, Y₂ with varied pH and i. Fig. 4b explains the effects of pH and i on Y₂. It can be observed in Fig. 4b that minimum Y₂ value is at pH ≈7.0. Y₂ value beyond pH ≈7.0 was observed increasing. This trend in color removal was observed at each i value, however, variation in changed Y₂ value is high at lower side of i. It can also be observed that increasing i value always increases Y₂ at each value of pH, and complete color removal is found at higher side of i. But, comparatively smaller i value is required at lower pH for complete color removal. Therefore, it may be concluded that lower i value is preferable at acidic pH range, while in basic pH range higher i value is required for the CBSOL LE red wool dye degradation and color removal.

Both the mechanism i.e. direct and indirect oxidation seems to be involved in the EO treatment of CBSOL LE red wool dye containing wastewater. Direct oxidation method of dye at anode involve the adsorption of generated ˙OH radicals in the oxide lattice of Ti/RUO₂ anode and subsequent oxidation of dye at the anode. The adsorption rate of ˙OH radicals has been reported to be high at highly acidic pH. Increasing current up to i ≈ 0.75 A in highly acidic pH, leads to increase the ˙OH radicals generation at the anode, which in turns oxidize the dye by direct oxidation method. Simultaneously, generated H₂O₂ and HO^˙₂ (eqn (4) and (5)) indirectly oxidise the organics. Moreover, in addition to this, increasing current also increases the rate of generation of the chloro-oxidant species HOCl in the bulk of solution, which dominates over all chloro-oxidant species (Cl₂, HOCl and ClO⁻) in the highly acidic pH, which indirectly oxidise the dye.

In highly basic pH of EO treatment of CBSOL LE red wool dye containing wastewater, due to the increase in current up to i ≈ 0.75 A, generated ˙OH radicals adsorption in the anode oxide lattice is decreased, and these generated ˙OH are transformed in to lower oxidation potential oxidants such as H₂O₂ and HO^˙₂ (eqn (4) and (5)) and help in oxidizing the dye via mediated oxidation. Due to this reason, the Y₁ and Y₂ was observed decreased up to pH value ≈8.0. Moreover, in highly basic pH, chloro-oxidant species ClO⁻, is in dominance. Oxidation potential of the ClO⁻ reported to be highest among all chlorine species (Cl₂, HOCl and ClO⁻).³³

Increasing current increases the concentration of ClO⁻ which subsequently increases the dye removal rate in addition to ˙OH radicals. Due to the formation of ClO³⁻ and ClO⁴⁻ in alkaline pH, ClO⁻ concentration is decreased. Therefore, the dye degradation/color removal is reduced in alkaline pH in comparison to the acidic pH.³⁵ Increasing current beyond 0.75 A in both the acidic and basic pH, promotes the evolution of O₂ by the electrolysis of water via eqn (6). This reduces the ˙OH radicals generation, and hence the dye/color removal is not improved by increasing the current beyond 0.75 A.

Fig. 4c and d shows effect of pH and t on Y₁ and Y₂. It can be seen that increasing t value up to ≈80 min always increases Y₁ and Y₂, and for all t > 80 min Y₁ and Y₂ was found to be increased marginally. This can be seen true at all values of pH. During EO treatment, the contaminants bond to the anode surface and grow like a film with the time. Therefore, due to the resistance of this film the performance of the EO process is affected, and hence, Y₁ and Y₂ values are limited. Complete color removal and dye degradation is observed in ≈80 min of treatment time at high acidic pH values; however, at higher pH, more than 90 min of treatment time is required for complete color and dye degradation.

Fig. 5a and b shows effect of pH, t and i on the energy consumed, Y₃. With increase in pH, it was found that Y₃ value always increases with i up to ≈0.75 A. For i > 0.75 A, Y₃ value was found to gradually decreased (Fig. 5a). Y₃ value is also affected with change in i value in same manner as with pH. Same trend was found with pH and t for energy consumed.

Fig. 5 Three dimensional response surface graph for the electro-oxidation of CBSOL LE red wool dye wastewater (a) % energy consumed versus pH and i (b) % energy consumed versus pH and t.

4.3. Optimization analysis

In this study, there are three responses Y₁, Y₂ and Y₃, and the optimum conditions for responses Y₁, Y₂ and Y₃ are not same. Furthermore, at optimum condition, Y₁ and Y₂ should be maximum while Y₃ minimum. Therefore, EO of dye wastewater was optimized in terms of maximization of responses Y₁ and Y₂ and simultaneous minimization of response Y₃ using the desirability function of CCD.⁴⁷

For this purpose some constraints for operational parameters were applied as shown in Table 4.

Table 4 Constraints applied for optimization process

Variables	Goal	Lower limit	Upper limit
pH	Is in range	4	10
t	Is in range	10	90
i	Is in range	0.25	1.25
% dye degradation	Maximize	30.82	97.77
% color removal	Maximize	50.85	98.98
Energy consumed	Minimize	0.602	7.283

---

Table 5 shows the optimum values of responses and parameters with corresponding desirability value for the individual and simultaneous (maximization of responses Y₁ and Y₂ and simultaneous minimization of response Y₃) optimization.

Table 5 Individual and simultaneous (maximization of Y₁ and Y₂ and minimization of Y₃) optimization results

Individual response optimization
Response	pH	t (min)	i (A)	Desirability
Y1 = 93.68%	4.0	45.98	0.99	0.97
Y2 = 100%	7.5	76.15	1.23	1.0
Y3 = 0.47 Wh	6.7	20.86	0.48	1.0
Simultaneous optimization of responses
Y1 = 91.75%	4.0	90	0.25	0.896
Y2 = 99%
Y3 = 2.327 Wh

---

The optimum values of parameters were found to be i = 0.25 A, t = 90 min and pH = 4.0, which produced overall desirability, D = 0.896. At this optimum condition, the Y₁, Y₂, and Y₃ suggested by CCD were to be 91.75%, 99.00% and 2.327 Wh, respectively. To verify the adequacy of the developed models with ANN, the confirmation simulation runs were carried out using optimized process parameters, which were within the simulation ranges defined earlier. Also, experiments were performed to confirm the values of responses Y₁, Y₂ and Y₃ at the optimum condition and found to be 89.87%, 96.71% and 2.029 Wh, respectively, which are close to the predicted values (Table 6).

Table 6 Comparison between the suggested and experimental values of responses at optimum condition

Responses	Predicted value	Experimental value
% dye degradation (Y1)	91.75%	89.87%
% color removal (Y2)	99.00%	96.71%
Energy consumed (Y3)	2.327 Wh	2.029 Wh

---

The optimum pH of treatment was found to be 4.0, therefore, it is interpreted that both the mechanisms of EO treatment i.e. direct and indirect oxidation are responsible for the dye degradation/color removal. In the EO reactor, due to electrolysis, water is oxidized leading to the formation of adsorbed ˙OH radicals, which directly oxidize the dye. Furthermore, these generated ˙OH are highly unstable and are rapidly transformed to H₂O₂ and HO^˙₂, which help in indirect oxidation of dye. Also, the presence of HOCl, which is dominating chloro-oxidant species (Cl₂, HOCl and ClO⁻) at pH = 4.0, oxidize the dye by indirect (mediated) method in addition to direct anodic oxidation by adsorbed ˙OH radicals. ˙OH radicals have the very high oxidation potential as compared to the chloro-oxidant species, HOCl. Therefore, the dye degradation/color removal is largely due to the direct anodic oxidation minutely by the HOCl at optimized condition. Since, experimental value of Y₁ (% dye degradation) and Y₂ (% color removal) were found to be 89.87% and 96.71%. Thus, even color is nearly completely removed, 10.13% of dye is present in the treated wastewater in degraded form.

5. Conclusion

Modeling of CBSOL LE red wool dye wastewater treatment by electro-oxidation method with Ti/RuO₂ anode was done successfully by ANNs, and optimized neurons for the process was concluded to be eight. The training versus target gives regression coefficient of 0.995 along with validation; test and all data sets regression coefficient value as 0.992, 0.996 and 0.995, respectively. This indicates that training of the ANN model was done accurately and model can be used to stimulate the outputs from a given inputs. Furthermore, The model suggested by CCD gives R² value of 0.9908, 0.9880 and 0.9659 for % dye degradation (Y₁), % color removal (Y₂) and energy consumed (Y₃) respectively. This advocates a good correlation between the stimulated data from ANNs and predicted values by CCD.

The optimized values of process parameters by CCD were found to be i = 0.25 A, t = 90 min and pH = 4.0, and at this optimum value, responses Y₁, Y₂ and Y₃ from the experiments conducted were found to be 89.87%, 96.71% and 2.029 Wh, respectively. It was found that both the mechanisms of EO treatment i.e. direct oxidation and indirect oxidation are responsible for the dye degradation/color removal. The removal was found to be largely due to the direct anodic oxidation, and minutely by the HOCl at optimized condition.

References

1. P. Gregory, Dyes Pigm., 1986, 7, 45–56.
2. C. Phalakornkule, S. Polgumhang, W. Tongdaung, B. Karakat and T. Nuyut, J. Environ. Manage., 2010, 91, 918–926.
3. C. O. Neill, F. R. Hawkes, D. L. Hawkes, N. D. L. Helena, M. Pinheiro and W. Delee, J. Chem. Technol. Biotechnol., 1999, 74, 1009.
4. http://www.colorbandindia.com/cbsol-le-reactive-dyes-for-wool.html, accessed on April, 2014.
5. D. Georgiou, P. Melidis and A. Aivasidis, Bioprocess Biosyst. Eng., 2002, 25, 79–83.
6. P. Cañizares, F. Martianez, C. Jimeanez, J. Lobato and M. A. Rodrigo, Environ. Sci. Technol., 2006, 40, 6418–6424.
7. W. Chen and J. Liu, Desalination, 2012, 285, 226–231.
8. J. P. Kushwaha, V. C. Srivastava and I. D. Mall, Water Res., 2010, 44, 5867.
9. G. Chen, Sep. Purif. Technol., 2004, 38, 11–41.
10. A. G. Vlyssides, M. Loizidou, P. K. Karlis, A. A. Zorpas and D. Papaioannou, J. Hazard. Mater. B, 1999, 70, 41–52.
11. L. Szpyrkowicz, C. Juzzolino, S. N. Kaul, S. Daniele and M. D. D. Faveri, Ind. Eng. Chem. Res., 2000, 39, 3241–3248.
12. T. H. Kim, P. Chulhwan, S. Eung-Bai and K. Sangyong, Desalination, 2003, 155, 59–65.
13. D. Dogan and H. Turkdemir, J. Chem. Technol. Biotechnol., 2005, 80, 916–923.
14. D. Rajkumar and J. G. Kim, J. Hazard. Mater. B, 2006, 136, 203–212.
15. D. Rajkumar, B. J. Song and J. G. Kim, Dyes Pigm., 2007, 72, 1–7.
16. N. Mohan, N. Balasubramanian and B. C. Ahmed, J. Hazard. Mater., 2007, 147, 644–651.
17. S. Raghu and B. C. Ahmed, J. Hazard. Mater. B, 2007, 139, 381–390.
18. S. Raghu, C. W. Lee, S. Chellammal, S. Palanichamy and B. C. Ahmed, J. Hazard. Mater., 2009, 171, 748–754.
19. R. Saravanathamizhan, N. Mohan, N. Balasubramanian, V. Ramamurthi and B. C. Ahmed, Clean: Soil, Air, Water, 2007, 35, 355–361.
20. G. B. Raju, M. T. Karuppiah, S. S. Latha, D. L. Priya, S. Parvathy and S. Prabhakar, Desalination, 2009, 249, 167–174.
21. A. I. del Río, J. Molina, J. Bonastre and F. Cases, Chemosphere, 2009, 75, 1329–1337.
22. Y. Kong, J. Yuan, Z. Wang, S. Yao and Z. Chen, Appl. Clay Sci., 2009, 46, 358–362.
23. W. Miled, A. Haj Said and S. Roudesli, J. Textile Apparel Technol. Manage., 2010, 6, 1–6.
24. Y. M. Zheng, R. F. Yunus, K. G. Nadeeshani Nanayakkara and J. P. Chen, Ind. Eng. Chem. Res., 2012, 51, 5953–5960.
25. S. Senthilkumara, B. C. Ahmed, M. Perumalsamy and H. J. Prabhu, Electrochim. Acta, 2012, 77, 171–178.
26. P. A. Carneiro, M. E. Osugi, C. S. Fugivara, N. Boralle, M. Furlan and M. V. B. Zanoni, Chemosphere, 2005, 59, 431–439.
27. C. Ramírez, A. Saldaña, B. Hernández, R. Acero, R. Guerra, S. Garcia-Segura, E. Brillas and J. M. Peralta-Hernández, J. Ind. Eng. Chem., 2013, 19, 571–579.
28. M. Diagne, V. K. Sharma, N. Oturan and A. Mehmet, Environ. Chem. Lett., 2014, 12, 219–224.
29. S. Singh, V. C. Srivastava and I. D. Mall, J. Phys. Chem. C, 2013, 117, 15229–15240.
30. S. Sundarapandiyan, R. Chandrasekar, B. Ramanaiah, S. Krishnan and P. Saravanan, J. Hazard. Mater., 2010, 180, 197–203.
31. C. Comninellis, Electrochim. Acta, 1994, 39, 1857–1862.
32. I. Sirés, E. Brillas, M. A. Oturan, M. Rodrigo and M. Panizza, Environ. Sci. Pollut. Res., 2014, 21, 8336–8367.
33. M. Deborde and U. Von Gunten, Water Res., 2008, 42, 13–51.
34. D. Pletcher and F. C. Walsh, Industrial Electrochemistry, Chapman and Hall, London, 1990.
35. Y. J. Jung, K. W. Baek, B. S. Oh and J. W. Kang, Water Res., 2010, 44, 5345–5355.
36. S. Lek and J. F. Guegan, Ecol. Modell., 1999, 120, 65–73.
37. M. S. Bhatti, D. Kapoor, R. K. Kalia, A. S. Reddy and A. K. Thukral, Desalination, 2011, 274, 74–80.
38. A. Aleboyeh, M. B. Kasiri, M. E. Olya and H. Aleboyeh, Dyes Pigm., 2008, 77, 288–294.
39. K. Hornik, M. Stinchcombe and H. White, Neural Network, 1989, 2, 359–366.
40. V. K. Pareek, M. P. Brungs, A. A. Adesina and R. Sharma, J. Photochem. Photobiol., A, 2002, 149, 139–146.
41. J. E. F. Moraes, C. A. O. Quina Nascimeto, D. N. Silva and O. Chiavone-Filho, Environ. Sci. Technol., 2004, 38, 1183–1187.
42. F. Despange and D. L. Massart, Analyst, 1998, 123, 157–178.
43. D. Salari, A. Niaei, A. Khataee and M. Zarei, J. Electroanal. Chem., 2009, 629, 117–125.
44. V. K. Sangal, V. Kumar and I. M. Mishra, Comput. Chem. Eng., 2012, 40, 33–40.
45. V. K. Sangal, V. Kumar and I. M. Mishra, Chem. Eng. Commun., 2014, 201, 72–87.
46. V. K. Sangal, V. Kumar and I. M. Mishra, Chem. Ind. Chem. Eng. Q., 2013, 19(1), 107–119.
47. J. P. Kushwaha, V. C. Srivastava and I. D. Mall, AIChE J., 2011, 57(9), 2589–2598.

---