Statistical modeling and optimization of heterogeneous Fenton-like removal of organic pollutant using fibrous catalysts: a full factorial design

This work focuses on the optimization of heterogeneous Fenton-like removal of organic pollutant (dye) from water using newly developed fibrous catalysts based on a full factorial experimental design. This study aims to approximate the feasibility of heterogeneous Fenton-like removal process and optionally make predictions from this approximation in a form of statistical modeling. The fibrous catalysts were prepared by dispersing zerovalent iron nanoparticles on polyester fabrics (PET) before and after incorporation of either polyamidoamine (PAMAM, –NH2) dendrimer, 3-(aminopropyl) triethoxysilane (APTES, –Si–NH2) or thioglycerol (SH). The individual effect of two main factors [pH (X1) and concentration of hydrogen peroxide-[H2O2]μl (X2)] and their interactional effects on the removal process was determined at 95% confidence level by an L27 design. The results indicated that increasing the pH over 5 decreases the dye removal efficiency whereas the rise in [H2O2]μl until equilibrium point increases it. The principal effect of the type of catalysts (PET–NH2–Fe, PET–Si–NH2–Fe, and PET–SH–Fe) did not show any statistical significance. The factorial experiments demonstrated the existence of a significant synergistic interaction effect between the pH and [H2O2]μl as expressed by the values of the coefficient of interactions and analysis of variance (ANOVA). Finally, the functionalization of the resultant fibrous catalysts was validated by electrokinetic and X-ray photoelectron spectroscopy analysis. The optimization made from this study are of great importance for rational design and scaling up of fibrous catalyst for green chemistry and environmental applications.

Ferrous ions consumed during the Fenton reaction are regenerated by further reaction of ferric ions with hydrogen peroxide, as in Eqs. 4-6 14 .
Although Fenton and Fenton-like treatments have noteworthy benefits [15][16][17][18] , homogeneous Fenton process containing free reagents (iron ions and hydrogen peroxide) have some critical disadvantages including selective reaction conditions (acidic pH), generation of a high amount of iron sludge and single-use reagent which is costly and counterproductive 19,20 . Therefore, much effort has been invested to improve the treatment efficiency of the Fenton reaction process as well as extending their working conditions and cutting down on secondary pollution 21,22 .
Heterogeneous Fenton process using various types of catalysts such as zero-valent iron (ZVI = Fe 0 ) [23][24][25][26] , goethite (-FeOOH) 27,28 , and Fe 3 O 4 [29][30][31] , has been extensively studied for pollutant degradation processes. Solid zerovalent iron allows a reduction in iron use, with the potential prospect to work in an operational pH. According to Xu et al. 32 , the zero-valent iron-based Fenton-like system provides better degradability of phenolic compounds even in working pH condition (pH 5-6). However, this process still produces iron sludge (due to the agglomeration of particles and limited recyclability of iron due to easy oxidation) as in homogeneous Fenton process. Hence, researchers introduce immobilization of zerovalent iron particles on solid supports to control the sludge formation and ensure reusability of the catalysts. Our recent studies reported the successful immobilization of zerovalent iron nanoparticles on high porosity support material (fibrous polyester fabric) and explored their effectiveness as a fibrous catalyst toward the removal of various organic and pathogenic contaminants in water through a heterogeneous Fenton process without generating secondary pollutants (iron sludge) 33 .
In general, the Fenton process is influenced by many factors including, the pH of the reaction medium, the temperature, the hydrogen peroxide concentration and the amount of catalyst which is needed to optimize this process for assessing their measurable performances 34 . Conventional optimization process was determined through the interaction of the individual factors at-a-time while keeping all other conditions constant, this approach does not take into account cross-effects from the factors considered and leads to a poor optimization result 35 . When a multifactor system is present, it is more appropriate to employ statistical-based optimization strategies to achieve such a goal 36 . An experimental design methodology is a modern approach, which has been widely used in several applications 1,37,38 also allowing the modeling of the process. In fact, the design of experiments (DoE) is used to identify or screen the important factors affecting a process or product and to develop statistically significant empirical models 35,39,40 . To our best knowledge, no study has reported the statistical modeling and optimization of heterogeneous Fenton-like removal of organic pollutant using fibrous catalysts through a full factorial design.
Therefore, in this study, an L 27 full factorial experimental design was used to optimize the removal process of organic pollutants (dye) by a heterogeneous Fenton-like process. Individual and interaction effects of pH and concentration of hydrogen peroxide [H 2 O 2 ] μl was studied in a form of statistical modeling. The responses considered were the reduction of dye concentration (%) after oxidation. Three different catalysts (based on the preparation process) chosen for this study were from our newly developed fibrous catalysts (PET-NH 2 -Fe|PET-Si-NH 2 -Fe|PET-SH-Fe) 20 . Electrokinetic and X-ray photoelectron spectroscopy (XPS) analysis of resultant materials were also studied. Synthesis and structural analysis of zerovalent iron nanoparticles, formation of Fe 2+ , reactive species and detailed mechanism of dye removal can be found in our previous studies 20,33 .

Experimental
Materials. Analytical grade polyamidoamine-PAMAM dendrimer, 3-(aminopropyl)triethoxysilane (APTES), 1-thioglycerol (SH), ammonium hydroxide (NH 4 OH); absolute ethanol (Et-OH); sodium borohydride (NaBH 4 ), crystal violet (CV), hydrogen peroxide (H 2 O 2 , 30% w/v) and iron nitrate (Fe(NO) 3 ) were purchased from Sigma Aldrich Ltd and used as received without any further purifications. Deionized water from Milli-Q ® Direct 8 water purification system was used throughout all experiments. The polyester nonwoven membrane used in this study was fabricated in The Swedish School of Textiles (University of Borås) based on a web of cylindrical polyethylene terephthalate fibres (average diameter 12 μm) formed by carding and consolidated by needle punching. The physical characteristics of the fabricated nonwoven are displayed in Table S1. where dl/dp is the slope of streaming current vs. differential pressure, η is electrolyte viscosity, ε represents the dielectric coefficient of electrolyte, ε 0 represents permittivity, L is the length of the streaming channel and A is the cross-section of the streaming channel.
X-ray photoelectron spectroscopy (XPS) study was carried out by a spectrophotometer (PHI 5500 ESCA, Physical Electronics INC., USA) equipped with monochromatic aluminium (Al) source (photon energy = 1,486.6 eV). Due to the insufficient conductivity, an electron neutralizer was used to compensate for the charge. Survey scan for the compositional evaluation (Energy range of 0-1,100 eV; Step size of 0.4 eV/step), and the narrow scan for the chemical state analysis with selected range for individual elements and step size of 0.1 eV/step was performed [41][42][43][44] . Morphological and structural analysis of such fibrous catalysts were extensively studied in our previous reports 20, 33 .
Heterogeneous Fenton-like removal of crystal violet dyes. The heterogeneous Fenton-like removal of crystal violet dye (50 mg/L) using a fibrous catalyst (25 mg/5 mL, PET-NH 2 -Fe, PET-Si-NH 2 -Fe, and PET-SH-Fe) and H 2 O 2 (100-500 μl/5 ml) was investigated in different pH (5,7,9) at 22 ± 2 °C. UV/visible spectrophotometer was used to monitor the removal process. Equilibrium time to attain the maximum dye removal has been detected individually for each catalyst. The minimum time required for complete removal of dye using a particular catalyst through an individual set of experiment was considered as equilibrium time. The pH range in this study was estimated based on our previous investigation on same catalysts 20 . The equilibrium concentration (Q e ) of crystal violet solution was measured by referring the absorbance calibration curve of the known standard solution, which allowed calculating the relative dye concentration (%) in the reaction chamber based on initial dye concentration. Based on the relative dye concentration in the chamber at the equilibrium point, the efficiency of the catalysts was estimated through a passive approach.

Design of experiment (DoE) and statistical modeling.
Matrix of the experiment's factors, their levels and response have been presented in Table 1. A full factorial design of experiment combining two factors According to the number of control parameters and their levels, a distinct full factorial design has been selected (L 27 ) to determine the optimal number of experiments in triplicates (See Table 2). In the end, a passive response from each trial run in terms of the relative concentration of dye at equilibrium was recorded. The availed results from each trial run are evaluated in "Minitab ® 17 graphical and statistical analysis tool" based on main effects, interaction, ANOVA, and a response table.

Results and discussions
The results of this study has been divided into two parts; (1) the first part will discuss the statistical modeling and optimization of the process parameters that affect the mean and variance of the attributes and identify the significant contribution and, (2) the second part will discuss the characterization of fibrous catalyst used in this study.

Part-1: Statistical modeling and optimization of the differences in the process parameters. Analysis of main effects.
The main effect is defined as the effect of an independent variable on a dependent variable averaged across the levels of all independent variable. By analyzing the main effects of the parameters, the general trends of the factor's influence on the process can be determined 45 . Figure 1 shows the main effect plots for the relative dye concentration (%) after Fenton-like oxidation using PET-NH 2 -Fe, PET-Si-NH 2 -Fe and PET-SH-Fe based on the date shown in Table 2. Results shown in Fig. 1a-c implies that, regardless of the type of the catalysts, with the changing in level in both X1 and X2, the response has significantly altered in a linear model which validates the statistically significant effect of pH and [H 2 O 2 ] μl on the response. Concentration may negatively affect the removal process as explained in various reports 8,20,46,47 .
Detailed analysis of the results shows that the lowest level of the factor X1 and highest levels of the factor X2 determined the maximum removal efficiency. With the increase in pH, the net negative surface charges on the catalysts surface cancelling the efficiency of reactive species that are responsible for the removal. Jung et al. (2013) reported that alkaline medium affects both reagents in the Fenton process causing instability in hydrogen peroxide and forming oxides on iron which significantly hinder the general of reactive species 22 . Lower concertation of hydrogen peroxide is too concentrated to react with mineralized iron to produce reactive species. However, www.nature.com/scientificreports/ it also needs to be noted that, after the saturation concentration of reactants, a further increase in concentration may negatively affect the removal process as explained in various reports 8,20,46,47 . Figure 2 shows the interaction effects of the factors in the low level and the high level of another factor. Figure 2a shows Fig. 2b,c). The overall scenario concludes that the removal efficiency increases with the increase in the concentration of hydrogen peroxide [H 2 O 2 ] at the same time decrease in pH also increases removal efficiency as consistent with main effects. This also indicates that the interaction plots are suitable for exploring the process parameters.

Analysis of interaction effects between factors.
Analysis of variance-ANOVA. The main effect and interaction plots provide the optimal levels of each factor but do not indicate which factor has the most significant impact on the output and the contribution of each factor. This was determined by the ANOVA, which is a robust method to determine the contribution of each factor and the significance of the optimization model 48 . It is accomplished by determining the Fischer's test value (F-value) and the sum of squares, which are used to evaluate the significance of the parameters; p values below 0.05 or 5% were considered statistically significant 49 . Tables 3, 4 Tables S2, S4, S6). On the other hand, by comparing the F-values among three fibrous catalysts considering all factors and levels, it can be seen that the type of catalyst does not have any significant impact on the dye removal. The 2-way interaction of the factors also shows an overall significant impact on dye removal. However, extensive analysis (see Tables S3, S5, S7) showing coefficient of each interaction identifies the interaction between X1 B and X2 A as not significant in PET-Si-NH 2 -Fe (see Table S5), X1 A and X2 B plus X1 B and X2 B in PET-SH-Fe (see Table S7). The influence of each factor is also expressed as the percent contribution (p%). From the results, it can be seen that pH has the percent contribution of 46.72% in PET-NH 2 -Fe, 48 ] μl can be explained by the synergic effect of pH on stabilization of hydrogen peroxide and formation of reactive species, which ultimately performs the removal of dyes, in an unlikely pH condition the no removal occurs, whereas in a favourable pH condition, lower or higher concentration of [H 2 O 2 ] μl will affect the removal but will not be absent 46,50 .    www.nature.com/scientificreports/ Fig. 3a, it is evident that the normal probability plot exhibits a nearly linear response, demonstrating that the errors are distributed normally. This is confirmed by the histogram, which also shows a normal distribution. The plot of the residuals versus the fitted values indicates that the residuals are randomly distributed around the zero lines, indicating that no sequential association exists and the errors have a constant variance. Moreover, the plot of the residuals versus the observation order is used to evaluate the pattern that may have an impact on the output. It is observed that the residuals are normally distributed close to the zero lines, which implies that there is a fair association with the parameters that do not require any further analysis of the errors 51-53 . A similar trend has been found in Fig. 3b,c.

Analysis of individual value plots.
Because individual value plots display all values for all groups at the same time, they are especially helpful when the study is dealing with variables, groups, and even subgroups. The individual value plot of dye concentration (%) after oxidation with PET-NH 2 -Fe, PET-SI-NH 2 -Fe and PET-SH-Fe fibrous catalysts, has been plotted for each factor and their levels (see Fig. 4 gives the best efficiency. On the other hand, in pH-7 the efficiency disrupted significantly by more than 50%. And finally, in the highest level of pH, there is no removal recorded regardless of the change in concentration of hydrogen peroxide. The same explanation corresponds to the other catalysts samples. These results provide significant statistical evidence to predict the best possible parameters and their levels.   The summary of the ζ-potential analysis of the samples is displayed in Fig. 6. It can be seen that the isoelectric point of untreated polyester fibres is pH = 3.9 which means that negative surface charge will be observed for the polyester fibres for pH higher than 3.9. Due to the further addition of -OH and -COOH groups during plasma treatment of the PET fibres (PET), the surface charge and isoelectric point were further negative starting from − 6.89 mV at pH = 3.45 to until − 49.23 mV at pH = 9.91 (see Fig. 6). With a negative surface containing -OH and -COOH groups may pose an ideal environment for robust electrostatic incorporation of cationic iron ions (Fe 3+ ) followed by reduction and immobilization into polyester fabrics. Upon incorporation of Fe 0 , the isoelectric point of PET-NH 2 -Fe, PET-Si-NH 2 -Fe and PET-SH-Fe shifted to pH around 7 signifies (see Fig. 6) an increase in ζ-potential which can be due to the loading of iron particles and reduction of oxygen-containing groups.

Analysis of fitted line plots.
X-ray photoelectron spectroscopy (XPS). The element composition and the nature of the chemical bonding of a plasma treated PET and PET-NH 2 -Fe were evaluated by XPS (see Fig. 7 and Fig. S2 of supplementary information). Results summarized in Table 6 shows that a control sample contains 76.4 a.t% carbon and 21.1 a.t% oxygen, which after plasma activation changed into 75.4 a.t% carbon and 24.6 a.t% oxygen indicating an increase in oxygen due to formation of -COOH and -OH groups on the fibre surface. After the modification of the fabric with PAM (PET-NH 2 ), nitrogen was detected in the surface composition having and iron was detected on PET-NH 2 -Fe sample. A detailed study of Fig. 7b (plasma-treated sample) illustrates that the sample consists mainly of C and O in a ratio around 3:1. Majority of the carbon (i.e. 60% of the area) refers to the C≡C and C=C bonding type in the hydrocarbon chain, whereas 40% of the area corresponds to C=O, C-O and COO − . O1s, on the other hand, shows 90% of the signal area is for C=O, C-O and -OH. All features in the O1s and part of C1s correspond to polyester, i.e. the COOfunctional group. Figure 7c and Fig. S2d shows the presence of 7.4 at.% nitrogen that confirms the grafting of PAM is mainly a graphitic-N type (R 3 N) and the rest refers to pyrrolic (R 2 NH). C-O proportion is enriched in C1s and O1s peaks. Figure 7d and Fig. S2f indicates that the sample consists of 5.0 at.% Fe in Fe(III) state. 2.3 at.% N at 400.0 eV corresponds to a 100% pyrrolic-N, secondary amine and/or primary amide condition, i.e. C 4 H 9 N/R 2 NH/RCONH 2 ). The C-O ratio is around 2:1. Alkane-type hydrocarbon chain, i.e. C1s = 284.8 eV (C-C) is determined. Rest of the C signal can be assigned to the carboxylic group, i.e. COO − . Moreover, most of the deconvoluted peaks in O1s also agrees with the features in a carboxylic group. In addition, a deconvoluted located at 529.5 eV corresponds to metal, could be metal-Fe (zerovalent iron).

conclusions
A full factorial design of experiment to predict and optimize the heterogeneous Fenton-like removal of organic pollutants using newly developed fibrous catalysts has been successfully achieved. The statistical insight on the effect of the factors (X1 and X2) and their levels in the removal process has been identified. Primarily, the statistical model showed that for both factors (X1 and X2) optimum values for the process variables exist. The results have been summarized below. a) Optimum operating conditions were determined based on a mathematical model in the form of statistical modeling, they are: initial acidic pH of 5, [H 2 O 2 ] μl of 500 μl which is a distinctive behaviour in the heterogeneous Fenton-like process. Under these conditions, 99% of pollutants are being removed in a rapid oxidation reaction. www.nature.com/scientificreports/ b) The interaction between X1 and X2 was of statistical significance with respect to the lower level of X1 and a higher level of X2 whereas the X1 turned into the key factor for the process. Whereas the type of catalysts turned out to be statistically insignificant. c) The ANOVA results provided evidence of the statistical significance of the factors and their corresponding levels that gives clear information during optimizing the process productivity and effective prediction. d) The ζ-potential and XPS analyses of the fibrous catalysts validated the surface modification of polyester fabric and loading of zerovalent iron nanoparticles.
While providing valuable insight into the roles of factors, the model provides rational ground to undertake a kinetic model that will further provide an in-depth understanding of the mechanisms and kinetics of the heterogeneous Fenton-like removal of complex pollutants in wastewater using such catalysts.   Table 6. Atomic proportion (a.t%) of the surface chemical composition of the samples.