Background

There are more than 10,000 chemicals used as colorants and among them, organic dyes are the most used and important ones. The properties of organic dyes, such as brightness, visible color at low concentrations, chemical structure, and resistance to light and chemical attack, make them fairly resistant to degradation [1]. Leather industry consumes widely organic dyes where large quantities of them remain in the effluent after dyeing and finishing processes. Different classes of dyes are used for different types of leather including the sole, upholstery, garment, gloves, and bags. But azo dyes come on the top of this list [2].

Due to the toxicity and slow degradation, the azo dyes are classified as environmentally hazardous materials [3]. These dyes are the major cause of both surface and ground water bodies contamination because of their carcinogenicity and toxicity to aquatic life, and they are also easily detected and with undesirable aesthetic aspects [4, 5].

A broad range of physicochemical methods has been proposed for treatment of dyeing effluents including biological [6], adsorption [7, 8], membrane [9], coagulation-flocculation [10], oxidation-ozonation [11], and advanced oxidation processes (AOPs) [12, 13].

Ozonation, one of the most used AOPs, is an oxidative process in which the oxidizing agent used is ozone (O3). Interest in the use of ozone in wastewater treatment has increased considerably in recent years due to the numerous advantages of this process. Among them is the high oxidation potential of ozone (2.08 V), even at low concentrations, with its high efficiency in the decomposition of organic matter and in the addition of oxygen to water [14]. In addition and unlike in other oxidizing agents such as Cl2, oxidation with O3 leaves no toxic residues that must be removed or disposed [15].

Degradation process with ozone proceeds in two different ways: direct and radical reaction. At acidic pH, molecular ozone selectively attacks specific parts of organic molecules like double bonds or aromatic systems of reactive dyes in direct reaction. At alkaline pH, hydroxide radicals coming from ozone decomposition react unspecifically with organic compounds [16]. The alkaline to neutral pH of dyeing wastewater favors the unspecific radical reaction. However, the pH of the solution may decrease around the dye molecules due to the formation of organic acids such as the ozonation products. Therefore, the reaction mechanism shifts towards the selective direct oxidation [17].

The mechanism of O3 reaction was studied by Hoigne [16]. Ozone reacts directly with pollutants via ozonolysis and through radical chain reaction. The reaction of ozone with hydroxide anions to intermediate radicals and to hydroxyl radicals is important for the oxidation of saturated organic compounds where no molecular ozonolysis is possible [18].

As it has been previously shown [19], the efficiency of color removal by ozone is dependent on various parameters such as the ozone dosage, initial dye concentration, initial pH, reaction time, and temperature of the solution. In conventional methods used to determine the influence of each one of these parameters, experiments were carried out by varying systematically the studied parameters and keeping constant the others. These should be repeated to all the influencing parameters, resulting in an unreliable number of experiments. In order to optimize the value of effective parameters with the minimum number of experiments, central composite design - the most widely used form of response surface methodology (RSM) - was employed to find improved or optimal process settings in an efficient use of the experimental data. Using RSM, it is possible to estimate linear, interaction, and quadratic effects of the factors and to provide a prediction model for the response.

This method has been used for modeling and optimization of decolorization process by advanced oxidation processes [20, 21]. However, to the best of our knowledge, decolorization of organic dye solutions by ozonation and the effect of interaction of effective parameters using central composite design have not been studied.

In this work, the central composite design has been applied to the modeling and optimization of decolorization of dye solutions containing acid black 1 (AB1), acid orange 7 (AO7), or acid yellow 19 (AY19) by ozonation have been studied. These three dyes are comprehensively used in Iran's tanneries for leather dyeing and finishing. The influencing factors (variables) investigated were the dye and ozone initial concentration, initial pH of the solution, and the temperature and decolorization efficiency was monitored as the process response. Counter plots and response surfaces were drawn to predict the efficiency of the decolorization process under different values of the independent parameters.

Methods

Reagents

Ozone was continuously produced by an ozone generator (model 1000 mg/L, Nano Paak Sayyal, Iran). The dyes, AB1, AO7, and AY19, were obtained from Azar Paints Company, Tabriz, Iran. Their structure and characteristics are given in Table 1. Sulfuric acid and sodium hydroxide were of laboratory reagent grade and were obtained from Merck Inc., Darmstadt, Germany.

Table 1 Chemical structure and characteristics of the dyes
Full size table

Ozonation experimental setup

For decolorization process, ozonation was carried out in a batch reactor, 500 ml in volume, equipped with a magnetic stirrer and a thermometer. Experimental setup for ozonation batch experiments is shown in Figure 1.

Figure 1
figure 1

Experimental setup for ozonation batch experiments.

Full size image

Experimental design

In this work, central composite design, which is a widely used form of RSM, was employed for the optimization of decolorization process. In order to evaluate the influence of operating parameters on the decolorization efficiency of the dyes, four independent factors were chosen: ozone concentration (X1), initial dye concentration (X2), initial pH of the solution (X3), and temperature (X4). A total of 31 experiments were done in this work, including 24 = 16 cube points, 2 × 4 = 8 axial points and seven replications at the center point.

Experimental data were analyzed using the response surface regression procedure of a statistical analysis system (NemrodW version 2000 LPRAI, Marseille, France). For statistical calculations, the variables X i were coded as x i according to the following relationship:

x i = X i − X 0 δX
(1)

where X0 is the value of X i at the center point and δX presents the step change [22]. The experimental ranges and the levels of the independent variables for dye removal are given in Table 2.

Table 2 Experimental ranges and levels of the independent test variables
Full size table

Results and discussion

Second-order polynomial model

In this work, a central composite design was used to propose a mathematical model of the process behavior. This method involves the performing of the experiments according to the design, factors, and levels selected, estimating the coefficients of the mathematical model to predict the response and finally, check the adequacy of the model obtained [23]. The central composite design (CCD) matrix and experimental results obtained during the ozonation process are presented in Table 3.

Table 3 The 4-factor central composite design matrix and the value of response function (CR%)
Full size table

A second-order polynomial response equation (Equation 2) was proposed to correlate the dependent and independent variables:

Y = b 0 + ∑ i = 1 k b i x i + ∑ i = 1 k b ii x i 2 + ∑ i = 1 k ∑ i ≠ j = 1 k b ij x i x j
(2)

where Y is a response variable of decolorization efficiency; b i , the regression coefficients for linear effects; b ii , the regression coefficients for quadratic effects; b ij , the regression coefficients for interaction effects; x i are coded experimental levels of the variables.

Based on the experimental results, the value of the coefficients was estimated and the second-order equation for predicting the efficiency of the process for each dye was expressed as follows (Equations 3, 4, and 5):

AB 1 Y = 74 . 286 + 1.083 x 1 – 7.083 x 2 + 6 . 250 x 3 + 2 . 333 x 4 – 1 . 250 x 1 x 2 − 0.750 x 1 x 3 + 0.000 x 1 x 4 + 2 . 250 x 2 x 3 + 0.250 x 2 x 4 + 0.750 x 3 x 4 – 1.009 x 1 2 + 0.491 x 2 2 – 1 . 259 x 3 2 – 0.134 x 4 2
(3)
AO 7 Y = 58.000 – 0.250 x 1 – 6 . 500 x 2 – 6 . 833 x 3 + 3 . 250 x 4 + 0.000 x 1 x 2 − 0.250 x 1 x 3 – 0.750 x 1 x 4 – 0.250 x 2 x 3 – 0.750 x 2 x 4 + 4 . 250 x 3 x 4 – 1 . 125 x 1 2 – 0.000 x 2 2 – 2 . 500 x 3 2 – 2 . 625 x 4 2
(4)
AY 19 Y = 83 . 429 + 1 . 250 x 1 – 7 . 750 x 2 + 10.500 x 3 – 2 . 417 x 4 – 1.000 x 1 x 2 – 1 . 125 x 1 x 3 + 0.125 x 1 x 4 + 4 . 375 x 2 x 3 – 3 . 625 x 2 x 4 – 0.500 x 3 x 4 + 0.455 x 1 2 – 1 . 795 x 2 2 – 4.045 x 3 2 – 0.170 x 4 2
(5)

The decolorization efficiencies (CR%) have been predicted by Equations 3, 4, and 5 and presented in Table 3. Based on the results indicated in this table, there are good agreements between the experimental and predicted values of decolorization efficiency.

Second-order polynomial models validation

The predicted values of decolorization obtained using models (Equations 3, 4, and 5) are plotted against the experimental results and shown in Figure 2. The coefficient of determination (R2) quantitatively evaluates the correlation between the experimental data and the predicted responses. This coefficient belongs to the line that best fits on the data of the scatter plot with equation Y = ax + b and it is obtained with regression analysis based on the minimization of the squared errors. When this factor is closer to 1 and also, the coefficients of the line equation are closer to 1 and 0, respectively, the model is better. As can be seen in this figure, the predicted values match the experimental results reasonably well with R2 of 0.936, 0.964, and 0.929 for AB1, AY19, and AO7.

Figure 2
figure 2

Comparison of experimental results and the predicted values of the response variable (CR%). Top image, AB1; middle, AY19; bottom, AO7.

Full size image

The values prove the good adjustments of the predicted values to the experimental results, since those indicate that 0.936, 0.964, and 0.929 values of the variability in the response could be explained by the models. These values imply 93.6%, 96.4%, and 92.9% of the variations for AB1, AY19, and AO7 dyes, respectively; removal efficiency are explained by the independent variables. This also means that these models do not explain only about 6.4%, 3.6%, and 7.1% of variations for the removal of AB1, AY19, and AO7, respectively.

Analysis of variance and fitting the quadratic model

Analysis of variance (ANOVA) is a mean to test the significance and adequacy of the model [24]. Table 4 indicates the results of the quadratic response surface model fitting in the form of analysis of variance. The values of R2 show that the proposed models predict the performance of the decolorization process with high accuracy. Adjusted R2 is also a measure of goodness of a fit, but it is more suitable for comparing models with different numbers of independent variables. It corrects the R2 value for the data counts and the number of terms in the model by using the degrees of freedom on its computations, so if there are many terms in a model and not very large sample size, adjusted R2 may be visibly smaller than R2[25, 26]. In this study, adjusted R2 values (0.895, 0.921, and 0.817 for AB1, AY19, and AO7 respectively) were very close to the corresponding R2 values (see Table 4).

Table 4 ANOVA for fit of the dye decolorization efficiency from central composite design
Full size table

ANOVA shows whether the variation from the model is significant or not when compared with the ones associated with residual error [27, 28]. The F value, which is the ratio between the mean square of the model and the residual error, performs this comparison. If the proposed model be a good predictor of the experimental results, the F value should be greater than the tabulated value of the F distribution for a certain number of degrees of freedom in the model at the level of significance α. The F values obtained, 19.21, 25.96, and 10.55 for AB1, AY19, and AO7, respectively, are clearly greater than the tabulated F (2.352 at 95% significance) confirming the adequacy of the model fits. The experimental error (1.98%, 1.90%, and 2.38% for AB1, AY19, and AO7 respectively) due to uncontrolled factors was calculated from the replicates of the center point.

The regression coefficient values, standard deviation, t value, and P value are given in Table 5. The Student’s t test was used to determine the significance of the regression coefficient of the parameters. The P values were also used as a tool to check the significance of each of the coefficients, which, in turn, are necessary to understand the pattern of the mutual interactions between the test variables. The larger the magnitude of t value and smaller P value, the more significant is the corresponding coefficient [29].

Table 5 Estimated regression coefficients and corresponding t and P values from the data of central composite design experiments
Full size table

b3 is significant at a level less than 5% for all the dyes studied. The interaction coefficient b23 for AB1 and AY19 and the interaction coefficient b34 for AO7 are also significant at a level less than 5%. Therefore, the initial pH for all the dyes (coefficient b3), the interaction of the dye concentration and initial pH for AB1 and AY19 removal (coefficient b23), and the interaction of initial pH and the temperature for AO7 removal (coefficients b34) are the most influential factors. The significance of these quadratic and interaction effects between the variables would have been lost if the experiments were carried out by conventional methods.

Moreover, the Pareto analysis is a mean that gives more significant information to interpret the results. It calculates the percentage effect of each factor on the response, according to the equation [30]:

Y = b 0 + ∑ i = 1 k b i x i + ∑ i = 1 k b ii x i 2 + ∑ i = 1 k ∑ i ≠ j = 1 k b ij x i x j
(6)

where P i is the relative importance of factor. Figure 3 shows the Pareto graphic analysis. As can be seen in this figure among the variables, initial dye concentration and pH of the solution have the more influential effect on the decolorization process efficiency.

Figure 3
figure 3

Pareto graphic analysis of the factors.

Full size image

So, the insignificant terms can be removed from the RSM models, and the reduced and best fitted models for decolorization efficiencies (CR%), developed from the regression procedure and supported by ANOVA and Pareto analysis, could be presented as (Equations 7, 8, and 9)

AB 1 Y = 74 . 286 + 1.083 x 1 – 7.083 x 2 + 6 . 250 x 3 + 2 . 333 x 4 – 1 . 250 x 1 x 2 + 2 . 250 x 2 x 3 – 1.009 x 1 2 – 1 . 259 x 3 2
(7)
AO 7 Y = 58.000 – 0.250 x 1 – 6 . 500 x 2 – 6 . 833 x 3 + 3 . 250 x 4 – 0.250 x 2 x 3 + 4 . 250 x 3 x 4 – 1 . 125 x 1 2 – 2 . 500 x 3 2 – 2 . 625 x 4 2
(8)
AY 19 Y = 83 . 429 + 1 . 250 x 1 – 7 . 750 x 2 + 10.500 x 3 – 2 . 417 x 4 – 1.000 x 1 x 2 – 1 . 125 x 1 x 3 + 4 . 375 x 2 x 3 – 3 . 625 x 2 x 4 – 0.500 x 3 x 4 – 1 . 795 x 2 2 – 4.045 x 3 2
(9)

Effect of variables as response surface and counter plots

The response surfaces were drawn with two variables kept constant at their zero level and the other two varying within the experimental ranges. These surface plots provide a method to predict the decolorization efficiency for different values of the tested variables.

In Figure 4, the response surfaces were developed as a function of temperature and initial pH, while the ozone concentration and the dye concentration were kept constant at 600 and 30 mg/L, respectively, being the central levels. The efficiency of the process increased with an increase in solution temperature, showing that the overall ozonation reaction is endothermic. This could be due to the increasing collision frequency of the dye molecules on the surface of the ozone bubbles as a result of the increasing temperature. On the other hand, the fraction of molecules that possesses energy in excess of activation energy also increases; consequently, the decolorization efficiency of the ozonation process increases accordingly. Lin and Lai [31] have also generalized a kinetic model consisting of multiple steps for ozonation and adsorption of dye molecules in the textile wastewater, where the increasing of temperature enhances the process efficiency.

Figure 4
figure 4

Response surfaces for decolorization process as a function of temperature and initial pH of the solution.

Full size image

The response surfaces also show that the ozonation process efficiency in alkaline pHs is the optimum. This could be explained by the fact that at alkaline pH, hydroxyl radicals coming from ozone decomposition react unspecifically with organic compounds. Lu et al. [32] have reported the same findings for treatment of wastewater containing azo dye by ozonation. They have mentioned that the removal rate of color is enhanced under the alkalinity condition due to the formation of hydroxyl radicals which has stronger oxidation ability.

Figure 5 shows the response surfaces of the decolorization efficiency as a function of initial dye concentration and the temperature of the solution while the two other variables were kept constant. Increasing the initial concentration of the dye provides more dye molecules in contact with oxidant species and consequently, the efficiency of the removal process increases. As the process efficiency did not decrease in high concentration of the dye (up to 50 mg/L), the amount of the oxidant (O3) must be enough. This means that the applied amount of ozone could oxidize even more concentrated dyes. The efficiency of the process increased with an increase in the solution temperature, and the reason of this trend was explained in the previous paragraph.

Figure 5
figure 5

Response surfaces for decolorization process as a function of initial dye concentration and temperature of the solution.

Full size image

Model optimization and confirmation

The value of independent variables was optimized to achieve the highest performance of the decolorization process. The desired goal in term of decolorization efficiency was defined as ‘maximize’ to reach the highest efficiency. The optimum values of the process variables for the maximum decolorization efficiency and the maximum values of decolorization that were verified experimentally are shown in Table 6. This shows that the strategy to optimize the decolorization process and to obtain the maximal efficiency by CCD for the decolorization of these organic dyes during ozonation process is successful.

Table 6 Decolorization efficiency at optimum values of the process parameters
Full size table

Experimental

The desired concentration of the dye with a defined pH value was fed into the Pyrex reactor. The pH of the solution was measured by pH meter (Hanna pH211, Hanna Instruments, Cluj-Napoca, Romania) and was not monitored during the experiment. During the injection of the ozone, agitation was done at maintained speed to keep the solution homogeneous. The concentration of ozone injected was determined by a flow meter. At different intervals proposed by the experimental design, 2-ml sample were taken and the remaining dye concentration was determined using Ultrospec 2000 (Biotech Pharmacia, UK, model 80-2106-00) UV–vis spectrophotometer, at wavelength of the maximum absorption and calibration curve. Using this method, the percent color removal could be obtained. The percent color removal (CR%) was expressed as the percentage ratio of decolorized dye concentration to that of the initial one.

Conclusions

It was found that response surface methodology could effectively predict and optimize the performance of ozonation process for decolorization of solution containing organic dyes, AB1, AY19, and AO7. High color removal (>75%) was obtained under optimal value of process parameters for dye solutions in the first 5 min of the removal process. Analysis of variance showed the high coefficient of determination values (R2 = 0.9365, 0.9643, and 0.9296 for AB1, AY19, and AO7 respectively), thus ensuring a satisfactory adjustment of the second-order regression model with the experimental data. The Pareto analysis of the model terms showed that the initial concentration of the dye and pH of the solution are the most influential variables. The variation of process efficiency as a function of independent variables was shown by graphical response surfaces. Also, the optimum values of the process variables for the maximum decolorization efficiency were calculated and then verified experimentally.