Turbidity and COD Removal from Municipal Wastewater Using a TiO₂ Photocatalyst—A Comparative Study of UV and Visible Light

Abstract

Water resources are depleting, and the availability and supply of clean, potable water are a global concern. Advanced oxidation processes (AOPs) possess immense prospects in water and wastewater treatment settings. This study investigated and optimized the photocatalytic treatment of wastewater using titanium dioxide (TiO₂) as the photocatalyst. The one-factor-at-a-time (OFAT) technique was employed to evaluate the effects of reaction time (20–100 min), mixing speed (20–100 rpm), and catalyst load (0.3–1.5 g/L) on pH, colour, turbidity, and chemical oxygen demand (COD) removal from actual municipal wastewater. Reaction time and catalyst load were then identified as the two key factors selected to be modeled and were optimized for turbidity and COD removal using the Central Composite Design (CCD) of response surface methodology (RSM). These statistical models were developed and used to optimize the operating conditions. The results obtained showed a desirability efficiency of 74.7% at a 95% confidence level. The RSM model predicted results at the optimum conditions and showed reasonable agreement with the experimental results obtained. The optimal responses achieved were 32.64% COD removal and 95.17% turbidity removal. A comparative study between UV light and visible light was also conducted at optimum conditions, whereby the UV light was demonstrated to be highly effective for turbidity and COD removal. The optimal responses achieved were 25.58% COD removal and 66.88% turbidity removal for visible light.

1. Introduction

The issue of water scarcity and how it can be solved continues to be a challenge to the world [1]. Herein, the continuous acceleration of water pollution due to population growth, urbanization, and industrialization, and the corresponding demand for non-renewable sources, has greatly increased [2]. Currently, South Africa’s municipalities are failing to provide clean water due to inadequate wastewater treatment facilities and limited or exhausted water resources. Piped water in two-thirds of South Africa’s municipalities do not meet minimum potable water standards [3,4,5]. The wastewater contains several contaminants, such as organic pollutants (dyes, phenolic compounds, surfactants, etc.), pathogenic microorganisms, and a range of different colloidal particles, which have harmful impacts on humans, animals, and the environment [6,7,8]. Their removal is therefore paramount. Several conventional wastewater treatment methods have been employed for decades. However, over the years, new and more challenging contaminants have emerged, and these contaminants have proven to be difficult to remove by traditional methods. Biological treatment methods commonly used are activated sludge and biofilm processes. However, high investment and operating costs, susceptibility to sludge swelling, and long pre-preparation cycles are the issues that exist in biological treatment processes [9]. Therefore, alternative and more advanced techniques are required [6,10]. Advanced oxidation processes (AOPs), an emerging wastewater treatment technology, displays great potential for the degradation of a variety of organic contaminants [2]. This process involves the generation of a highly reactive hydroxyl radical, which destroys the contaminants. Photocatalytic oxidation is a type of AOP whereby the photo-activated reactions are distinguished by the free radical mechanism initiated by the interaction of photons of an appropriate energy level with the catalyst [11].

Several researchers have investigated the use of photocatalysis for targeting individual pollutants. Most antibiotics are metabolized partially in the body, but 30% to 90% of them are discharged into the municipal sewage system without being metabolized [12,13]. Okhovat et al. [14] reported on the application of a photocatalytic process using nanoparticles for TiO₂ on the removal of COD and metronidazole. The results on real wastewater (containing metronidazole at 80 mg/L and COD equal to 900 mg/L) showed that metronidazole and COD removal efficiencies at optimal conditions were 58.32% and 34.32%, respectively. Tayade et al. [15] found that an increase in catalyst load increased the decolourization and degradation percentages in their work to degrade methylene blue dye. However, the catalyst load may not exceed a certain quantity; otherwise, the degradation and decolourization percentages start to decrease due to light scattering and screening effects. This causes the catalyst’s surface to become unavailable for photon absorption. Furthermore, refs. [16,17,18,19] reported that excessive catalyst loading results in turbidity and a blocking effect that reduces light transmission through the whole solution and causes light scattering and a screening effect that decreases the degradation and decolourization percentages. The effect of light intensity is vital in the photocatalytic process, as the presence of sufficient light energy and wavelength allows for photocatalyst activation and enhances the generation of the hydroxyl and oxide radicals. In addition, refs. [16,17,20] stated that to achieve a high photocatalytic rate, specifically in water treatment, a reasonably high light intensity is required to sufficiently provide each TiO₂ active site with enough photon energy. In addition to light intensity, light wavelength also affects the efficiency of photocatalytic degradation, especially when using common photocatalysts such as ZnO (zinc oxide), SnO₂ (tin oxide), and TiO₂, as they are only active when exposed to UV light, which accounts for only 5% of the solar spectrum or 388 nm [21]. As a result of their large bandgap (Eg > 3.0 eV), most photocatalysts are unable to take advantage of visible and infrared light, which accounts for 43% and 52% of solar energy, respectively. Additionally, refs. [22,23,24] stated that using commercial Degussa P-25, which has a crystalline ratio of anatase to rutile of 70/80:20/30 and a light wavelength of less than 380 nm, is adequate for photon activation. UV light provides a wavelength that ranges from 315 nm to 400 nm and therefore provides sufficient light photons for the photonic activation of the catalyst. Additionally, refs. [25,26,27,28,29] studied the effect of reaction time on the degradation efficiency of organic pollutants and found that an increase in reaction time increased the degradation efficiency, as the highest degradation efficiency was correlated to the longest reaction time. This was attributed to an increased interaction time between the pollutant and the surface of the photocatalyst. Farouq et al. [30] studied the effect of the mixing speed on the photocatalytic degradation of aqueous ammonia and found that an increase in mixing speed led to a higher percentage of ammonia removal. In addition, Yin et al. [31] also studied the effect of mixing speed with respect to decolourization and determined that when the mixing speed was increased, the bulk solution improved its transfer to the catalyst’s surface, therefore increasing decolourization.

The optimization of process parameters in the photocatalytic degradation process is complex and may require many experimental results. However, the Response Surface Methodology (RSM) employed in this study has the ability to produce a large amount of information from a limited number of experiments [32]. The numerical optimization technique will be utilized to ascertain the optimum conditions and interactions between the operating parameters.

The application of photocatalysis concerning the treatment of actual industrial wastewater is very limited [33]. In addition, the study explored the different light sources (UV and UV-visible) for the photocatalytic degradation of wastewater. In this study, the efficiency of TiO₂ was studied and analyzed based on the removal of turbidity and chemical oxygen demand (COD) from actual municipal wastewater in a photo-catalytic system. Three operating parameters were varied: reaction time (20–100 min), mixing speed (20–100 rpm), and catalyst load (0.3–1.5 g/L), and their effect on pH, colour, turbidity, and COD removal from the wastewater was analyzed. RSM was used to optimize and study the interactions between the operating parameters using simulated wastewater by generating the values of the reaction time and catalyst load that produced the best (COD) removal and turbidity removal. These optimized parameters were then applied to UV/UV-Vis light sources, and a comparative study between the two light sources was performed to determine which light source achieved better results for COD and turbidity removal. The advantage of visible-light photocatalysis lies in its use of clean, renewable, and cheap visible light as a driving force. The sampled wastewater effluent used for the experiment was obtained from a local South African Municipal wastewater treatment plant.

2. Materials and Methods

2.1. Chemicals Used

CaCl₂H₂O, Peptone, Glucose, NaHCO₃, Urea, Meat extract, MgSO₄, K₂HPO₄, CuCl₂.7H₂O, NaCl and Titanium (IV) oxide were the chemicals used in the study. The characteristics of the commercial titanium dioxide used are as follows: the ratio of anatase to rutile is 75/25, particle size is 1–150 nm, density is 4.6 g/mol, and purity is 99%. All chemicals were supplied by Sigma Aldrich, Durban, South Africa.

2.2. Effluent Sample and Analytical Methods

2.2.1. Effluent Sample

Municipal wastewater from a local eThekwini municipality wastewater treatment plant based in Kwa-Zulu Natal, South Africa, was used. This wastewater was characterized and had the following characteristics: pH (6.85), colour (878 Pt.Co), turbidity (365 NTU), and COD (432 mg/L). The COD value was on the low side when the sample was taken. This municipality’s water was used for one factor at a time analysis. However, due to the limited availability of wastewater in the laboratory and the lack of consistency from the municipality’s treatment plant, synthetic wastewater was also simulated using analytical-grade chemicals that were mixed into a solution of 20 L of distilled water and 5 L of municipal wastewater. The composition of the chemicals used was adapted from Sibiya et al. [34] and is presented in

Table 1. Chemical composition of synthetic wastewater [34].

Item No	Chemical	Mass Added (g)
1	CaCl2H2O	1
2	Peptone	40
3	Glucose	27.52
4	NaHCO3	68.75
5	Urea	7.52
6	Meat extract	2.5
7	MgSO4	0.5
8	K2HPO4	7
9	CuCl2.7H2O	0.0125
10	NaCl	2.2

. The synthetic wastewater was characterized for turbidity (317 NTU) and COD (870 mg/L). The COD value for synthetic water is higher than for the sample taken, which represents typical values for the treatment plant. This wastewater was used for RSM studies.

2.2.2. Analytical Methods

The pH and the turbidity were analyzed using the pH meter HI98130 and the turbidity meter HI98703-02, respectively (HANNA instruments). The COD and colour were analyzed by Spectrophotometer DR 3900 (HACH), using the stored programs 435- COD HR and 125- colour 465 nm. The COD and colour removal percentages were determined by using Equations (1) and (2), with the same equation set up applicable to turbidity percentage removal:

$$\mathrm{COD} \mathrm{removal} \%=\frac{C_i-C_f }{C_i} \times 100$$

(1)

where $C_i$ and $C_f$ are the initial and the final COD concentrations (mg/L) before and after treatment, respectively [35].

$$\mathrm{Colour} \mathrm{removal} \%=\frac{C{l}_i-C{l}_f}{C{l}_i} \times 100$$

(2)

where $C{l}_i$ and $C{l}_f$ are the initial and the final Colour (Pt.Co) before and after treatment, respectively [36].

2.3. Experimental Setup

Figure 1 shows the set-up of a laboratory-scale photochemical reactor (Lelesil Innovative Systems). The photochemical reactor consists of a reaction vessel that has a 1.5 L capacity and an immersion well made of quartz, which houses the UV lamp. A cold-water circulating tank is used to cool down the immersion well that contains the lamp. A 250 W, 365 nm mercury UV and UV-Vis lamp was used.

As per the experimental design, experiments were performed based on one factor at a time. The first experiment varied reaction time from 20–100 min in increments of 20 min, whilst reaction speed and catalytic load remained constant at 60 rpm and 0.9 g/L, respectively. The second experiment varied mixing speed from 20–100 rpm in increments of 20 rpm, whilst reaction time and catalytic load remained constant at 30 min and 0.9 g/L, respectively. The third experiment varied catalyst load from 0.3–1.5 g/L in increments of 0.3 g/L, whilst reaction time and mixing remained constant at 30 min and 60 rpm, respectively. For each of the experiments, the reaction mixture was prepared by adding known amounts of TiO₂ to municipal wastewater that was filled to just below the sample port of the reaction vessel, and samples were collected after a 10-min settling time was implemented after each increment. A similar set of experiments was performed; however, this time the input parameters generated by response surface methodology on design experts were used.

2.4. Response Surface Methodology (RSM)

Response surface methodology (RSM) was used in this study. RSM is a useful optimization tool for developing and studying the impact of independent and interactive factors in a process. These include process optimization, enhancement, and the development of products [25]. Essentially, RSM establishes the relationship between a variable and its effect (response), with the advantage of producing a mathematical model. The design of experiments was performed based on the central composite design (CCD) adopted from RSM using Design Expert (version 13.0.7) developed by Stat-Ease, Minneapolis, MN, USA. There were 2 numeric factors that generated 5 center points and 13 experimental runs (which were generated randomly). The two factors, reaction time and catalyst load, were assessed for two responses, turbidity removal and COD removal, and to the best of our knowledge, this is the first analysis that has been reported for municipality wastewater using these responses.

3. Results and Discussion

3.1. Characterization of Municipal Wastewater

The municipal wastewater properties are as follows: pH (6.85), colour (878 Pt.Co), turbidity (365 NTU), and COD (432 mg/L), as stated earlier, and these values were used as the basis for the analysis of the experimental results obtained.

3.2. Effect of Reaction Time on Photocatalysis Treatment

For this experiment, the effect of reaction time on the output parameters was studied. The reaction time was varied from 20 min to 100 min, while the mixing speed and catalyst load were kept constant at 60 rpm and 0.9 g/L, respectively. The reaction time had minimal effect on the pH (Figure 2a), as the pH values remained almost constant, ranging from 6.79 to 7.11, further validating the minimal effect reaction time had on pH, as the pH also remained close to the characterized wastewater pH of 6.85. Constant pH suggests that the products of photocatalysis had no significant influence on pH.

The removal efficiencies of the parameters colour (Figure 2b), turbidity (Figure 2c), and COD (Figure 2d) were highest in the first 20 min and then decreased thereafter. The initial increase is because, with an increase in contact time, the availability of hydroxyl radicals for the oxidation of pollutants present in wastewater increases [35]. Kumar and Pandey [29] studied the effect of reaction time on the photodegradation of methyl green and also found that an increase in reaction time increased the degradation efficiency, which they attributed to increased interaction time between the pollutant and the surface of the photocatalyst. The decrease thereafter might be because the TiO₂ photocatalyst aggregated with the pollutants, and with time, the interactive surface of the photocatalyst became saturated and dissociated, contributing to secondary complex pollutants [15,16].

3.3. Effect of Mixing Speed on Photocatalysis Treatment

For this experiment, the effect of mixing speed on the output parameters was studied. The mixing speed was varied from 20 rpm to 100 rpm while the reaction time and catalyst load were kept constant at 30 min and 0.9 g/L, respectively. The mixing speed had minimal effect on the pH (Figure 3a), as the values remained constant, ranging from 6.77 to 7.39. Farouq et al. [30] studied the effect of the mixing speed on the photocatalytic degradation of aqueous ammonia and found that an increasing mixing speed leads to a higher percentage of ammonia removal. In contrast to this, as the mixing speed increased, the removal efficiencies for colour (Figure 3b), turbidity (Figure 3c), and COD (Figure 3d) decreased. So it is believed that as stirring speed is increased, TiO₂ particles disintegrate, contributing to an increase in the three parameters [37].

3.4. Effect of Catalyst Load on Photocatalysis Treatment

For this experiment, the effect of catalyst load on the output parameters was studied. The catalyst load was varied from 0.3 g/L to 1.5 g/L, while the reaction time and mixing speed were kept constant at 30 min and 60 rpm, respectively. The catalyst load had a minimal effect on the pH (Figure 4a), as no significant change in pH was observed; it ranged from 6.82 to 7.24. The catalyst load removal efficiency results for colour (Figure 4b), turbidity (Figure 4c), and COD (Figure 4d) decreased with an increase in catalyst load. These results concur with the findings of previous researchers [15,19], who reported that removal efficiencies decreased above the optimum catalyst load. Mecha et al. [38] investigated the effect of catalyst concentrations (0.1, 0.5, and 1.0 g/L) and reported that the percentage removal of phenol increased with catalyst loading until 0.5 g/L and thereafter decreased at higher catalyst loading. Furthermore, refs. [16,17,19] found that excessive catalyst loading results in turbidity and a blocking effect that reduces light transmission through the whole solution and causes light scattering and a screening effect that decreases the degradation and decolourization percentages.

It was observed that reaction time and catalyst load, turbidity, and COD removal were the most significant input and output variables, respectively, and therefore these were the parameters used for the optimization. Furthermore, the optimal ranges for catalyst load and reaction time were determined to be 0.3–0.6 g/L and 20–40 min, respectively. These conditions were applied to the optimization.

3.5. Previous Similar Work

The TiO₂ treatment achieved high removal efficiencies for turbidity and colour removal. A comparison with other actual wastewater studies from different wastewater sources shows that the removal efficiency for COD was very low in this work (

Table 2. Comparative studies on TiO₂ photo-catalytic degradation of actual wastewater.

Catalyst	Wastewater	Initial Conditions	Removal Efficiency	References
TiO2	Municipal Sewage	COD: 620 mg/L	12% 76% 80% pH: 7.1	This work
		Colour: 878 Pt.Co
		Turbidity: 365 NTU
		pH: 6.85
		Cat. Loading: 0.1 g/L
		Reaction time 150 min
TiO2	Real greywater	COD: 620 mg/L	54% pH: 4.45	[41]
		pH: 5
		Cat. Loading: 0.1 g/L
		Reaction time 150 min
TiO2	Petroleum refinery	COD: 1226 mg/L	92%	[42]
		pH: 8
		Cat. Loading: 1.5 g/L
		Reaction time: 150 min
TiO2	Petroleum refinery	COD: 8200 mg/L	60%	[43]
		pH: 4.5
		Cat. Loading: 1 g/L
TiO2	Paper mill	COD: 2075 mg/L	75%	[44]
		pH: 6.5
		Cat. Loading: 0.75 g/L
		Reaction time 180 min
TiO2/calcium aluminosilicate	Municipal Sewage	COD: 2487 mg/L	94%	[28]
		pH: 6.5
		Cat. Loading: 0.75 g/L
		Reaction time: 180 min

). The removal of COD is an important indicator for evaluating the mineralization degree of the system [39]. The low activity in COD removal in this work will suggest that TiO₂ usage for treating eThekwini municipality wastewater could face challenges if used as a standalone technology. The difference in removal efficiencies between this work and the literature may be due to the difference in the wastewater composition. This will lead to different potential scavengers in the systems, and therefore, researchers are encouraged to do a more detailed analysis of the wastewater to include potential scavengers. Another potential cause of low activity is the number of ions dissolved in the wastewater (ionic strength). In the presence of strong electrolytes, the solubility of oxygen is lower than in pure water [40], hence lower activity.

3.6. Response Surface Modelling and Optimization

3.6.1. Turbidity

Synthetic wastewater was characterized and used for the RSM experimental runs. Turbidity and COD were 317 NTU and 870 mg/L.

Table 3. Operating parameters generated from RSM and the experimental responses.

		Factor 1	Factor 2	Response 1	Response 2
Std	Run	A: Catalyst Load (g/L)	B: Reaction Time (min)	Turbidity Removal (%)	COD Removal (%)
4	1	0.6	40	95.46	42.41
3	2	0.3	40	96.99	4.02
13	3	0.45	30	79.5	23.45
8	4	0.45	40	86.56	21.38
10	5	0.45	30	85.96	31.95
12	6	0.45	30	85.99	44.25
11	7	0.45	30	93.91	46.44
1	8	0.3	20	99.14	1.84
6	9	0.6	30	99.24	26.55
7	10	0.45	20	85.77	31.03
9	11	0.45	30	87.79	52.64
2	12	0.6	20	96.15	25.63
5	13	0.3	30	97.92	3.1

shows the operating parameters (factor 1 and factor 2) generated by RSM. Once these runs were conducted, the results for the output parameters were entered. Next, the two responses were then analyzed by the Design Expert.

The fit summary for response 1, shown in

Table 4. Fit summary for Turbidity response model.

Source	Sequential p-Value	Lack of Fit p-Value	Adjusted R2	Predicted R2
Linear	0.9766	0.1954	−0.1943	−0.7432
2FI	0.9246	0.1536	−0.3256	−2.0736
Quadratic	0.0053	0.9639	0.6196	0.5647	Suggested
Cubic	0.8744	0.8583	0.4952	0.4763	Aliased

, was generated, and the quadratic model was suggested based on the p-value; see Equation (3).

The results for the analysis of variance (ANOVA) for the quadratic model can be seen in

Table 5. ANOVA for quadratic model for response 1: turbidity.

Source	Sum of Squares	Degree of Freedom	Mean Square	F-Value	p-Value
Model	395.84	5	79.17	4.91	0.0301	significant
A-Catalyst load	1.71	1	1.71	0.1058	0.7545
B-Reaction time	0.7004	1	0.7004	0.0434	0.8409
AB	0.5329	1	0.5329	0.0330	0.8609
A2	357.70	1	357.70	22.18	0.0022
B2	2.96	1	2.96	0.1833	0.6814
Residual	112.90	7	16.13
Lack of Fit	6.86	3	2.29	0.0863	0.9639
Pure Error	106.04	4	26.51
Cor Total	508.74	12				not significant
R2 0.7781	Adjusted R2 0.6196	C.V.% 4.39	Predicted R2 0.5647	Adeq. Pr 4.8711	Mean 91.57	SD 4.02

, and the model F-value of 4.91 indicates the model is significant. It also suggests that an F-value this large only has a 3.01% chance of occurring due to noise. The p-value generated is 0.0301 and indicates that the model term is significant as it is less than 0.0500. Specific to this case, A² is a significant model term. Any p-value that is greater than 0.1000 suggests the model terms are not significant; however, hierarchy terms are not included in the terms counted. Therefore, this model is significant and acceptable as only model terms AB and B² are above 0.100. Lastly, the value of 0.9639 for the lack of fit indicates that it is not significant relative to pure error, and a non-significant lack of fit value indicates that the model will fit. The predicted R² value of 0.5647 is in reasonable agreement with the adjusted R² value of 0.6196, as the difference between the two values is less than 0.2. The adequate precision (Adeq. Pr) of 4.8711 shows an adequate signal as the value is greater than 4, which is desirable.

The predicted values were plotted against the experimental values using an experimental predicted data interactive plot to approximate the extent of the correlation. Figure 5a depicts a reasonably strong linear correlation between the experimental and predicted data with a high regression coefficient (Table 5), with only a couple of data points as outliers. The standard error (SE) for the line of best fit showed an insignificant deviation, as the p-value was less than 0.05 at a 95% confidence level. The modified RSM and Central Composite Design (CCD) (RSM-CCD) were used to demonstrate the interactive impact of the input parameters (factors) on the output parameters (response). Figure 5b displays the presence of interactions between the factors, catalyst load, and reaction time on the turbidity response. The graphical demonstration of the three-dimensional (3D) surface plot of the response model was selected based on the interaction of the significant factors that play a role in maximizing the desirability of the system. A curvature of considerable magnitude can be seen in Figure 5b. This curve also suggests that the correlation between the factors (AB) and the response (turbidity) were well-fitted on a quadratic function (Equation (3)). Graphically (Figure 5b), it was demonstrated that the response surface showed an arc with the optimal region for turbidity at the high-low levels of A (catalyst load), whilst the turbidity removal remained constant along the varying reaction times and therefore did not have as significant an effect as the catalyst load.

Turbidity = 86.79 − 0.5333A − 0.3417B + 0.3650AB + 11.38A² − 1.03B²

(3)

3.6.2. COD

The fit summary for response 2, shown in

Table 6. Fit summary for COD response model.

Source	Sequential p-Value	Lack of Fit p-Value	Adjusted R2	Predicted R2
Linear	0.1033	0.2772	0.2379	−0.0206
2FI	0.6440	0.2311	0.1742	−0.5861
Quadratic	0.0522	0.5445	0.5432	−0.0752	Suggested
Cubic	0.6532	0.2932	0.4607	−6.2294	Aliased

, was generated, and the quadratic model was suggested in Equation (4).

The results for the analysis of variance (ANOVA) for the quadratic model can be seen in

Table 7. ANOVA for quadratic model.

Source	Sum of Squares	Degree of Freedom	Mean Square	F-Value	p-Value
Model	2356.51	3	785.50	6.85	0.0106	significant
A-Catalyst load	1222.08	1	1222.08	10.66	0.0098
B-Reaction time	14.45	1	14.45	0.1260	0.7308
A2	1119.98	1	1119.98	9.77	0.0122
Residual	1032.12	9	114.68
Lack of Fit	474.44	5	94.89	0.6806	0.6632	not significant
Pure Error	557.69	4	139.42
Cor Total	3388.63	12
R2 0.6954	Adjusted R2 0.5939	C.V.% 39.25	Predicted R2 0.4378	Adeq. Pr 6.0593	Mean 27.28	SD 10.71

, and they suggest that the Model F-value of 6.85 indicates the model is significant. It also suggests that an F-value this large only has a 1.06% chance of occurring due to noise. The p-value generated is 0.0106 and indicates that the model term is significant as it is less than 0.0500. Any p-value that is greater than 0.1000 suggests the model terms are not significant; however, hierarchy terms are not included in the terms counted. Therefore, this model is significant and acceptable, as only model term B is above 0.100. Lastly, the value of 0.6632 for the lack of fit indicates that it is not significant relative to pure error, and a non-significant lack of fit value indicates that the model will fit. Table 7 indicates that the predicted R² value of 0.4378 is in reasonable agreement with the adjusted R² value of 0.5939, as the difference between the two values is less than 0.2. The Adeq. Pr of 6.0593 shows an adequate signal as the value is greater than 4, which is desirable.

The predicted values were plotted against the experimental values using an experimental predicted data interactive plot to approximate the extent of the correlation. Figure 6a shows that only a few data points were recurrently scattered around the line of best fit, which could explain the inconsistent COD % removal (organic degradation) based on the specified experimental conditions. The standard error for the line of best fit showed an insignificant deviation with a p-value less than 0.05 at a 95% confidence level. The COD model regression is low (Table 7) despite the model’s adjusted and predicted R² values being in reasonable agreement with a difference of less than 0.2. A curvature of considerable magnitude can be seen in Figure 6b. This curve also suggests that the correlation between the factors (AB) and the response (COD) was well fitted to a quadratic function (4). Graphically (Figure 6b), it was demonstrated that the response surface showed an arc with the optimal region for turbidity at the higher levels of A (catalyst load), ranging from 0.48 g/L to 0.54 g/L, whilst the COD removal remained constant along the varying reaction time and therefore did not have as significant an effect as the catalyst load.

COD = 35.88 + 14.27𝐴 + 1.55𝐵 − 18.62𝐴²

(4)

3.6.3. Optimization Using RSM

To optimize the experimental design, constraints (

Table 8. Constraints.

Name	Goal	Lower Limit	Upper Limit	Lower Weight	Upper Weight	Importance
A: Catalyst load	maximize	0.3	0.6	1	1	3
B: Reaction time	maximize	20	40	1	1	3
Turbidity	maximize	79.5	99.24	1	1	3
COD	maximize	1.84	52.64	1	1	3

) had to be implemented. Specific to this case, all input and output parameters (factors and responses) were maximized. This generated 8 solutions (

Table 9. Solutions generated.

Number	Catalyst Load	Reaction Time	Turbidity	COD	Desirability	Desirability (w/o Intervals)
1	0.600	40.000	96.628	33.082	0.747	0.855	Selected
2	0.600	39.580	96.712	33.017	0.746	0.851
3	0.598	40.000	96.380	33.334	0.746	0.852
4	0.600	39.387	96.750	32.987	0.745	0.849
5	0.600	38.327	96.942	32.822	0.741	0.838
6	0.586	40.000	94.622	35.060	0.734	0.831
7	0.600	36.449	97.224	32.531	0.727	0.817
8	0.600	34.317	97.457	32.200	0.706	0.790

), and one was selected as the optimal solution (Figure 7) as it presented the highest desirability of 74.7% removal efficiency. The optimal solution suggests a turbidity and COD removal of 96.6% and 33.1%, respectively, were attained at a catalyst load of 0.6 g/L and a reaction time of 40 min.

3.6.4. Validation of Optimized Conditions

The selected optimal conditions were validated and confirmed experimentally (

Table 10. Modified RSM-CCD optimum conditions experimental validation.

Response	Predicted	Actual	Difference
Turbidity (%)	96.63	95.17	1.46
COD (%)	33.08	32.64	0.44

) and were in good agreement with the predicted values, as the difference between the predicted and actual values was minimal (Figure 8a,b). This suggests the model’s predictability was consistent (p < 0.05) at 95% confidence levels.

It was also determined, according to the South African National Standards (SANS) 241 and the South African wastewater and industrial effluent law (2013), that the COD output parameter (498 mg/L) met the standards, whilst the turbidity output parameter (15.3 NTU) did not. This might be due to the syringe filter not being able to completely remove all TiO₂ nanoparticles. We observed a low value for COD removal, and these results are consistent with what was observed by Okhovat et al. [14], who observed a 34.32% COD removal. The potential reasons for a low COD removal are discussed in Section 3.5, and this could be due to scavengers and the high ionic strength of the wastewater.

3.7. Comparative Study between UV and Vis Light

Finally, once the optimal operating parameters were verified, a comparative study was performed, whereby the same experiment was conducted using UV-Vis light. The turbidity and COD removal percentages obtained under UV-Vis irradiation are lower than those obtained under UV irradiation. The optimal responses achieved for this were 25.58% COD removal and 66.88% turbidity removal. This is possibly due to TiO₂ not being photoexcited by visible light due to its large energy band gap [22], and this is in line with the literature [45]. Due to turbidity caused by TiO₂ particles, additional post-treatment separation methods are required. Further studies making use of TiO₂ as a catalyst for photocatalytic wastewater treatment should include additional post-treatment filtration steps to recover the TiO₂ to ensure optimal removal of the nanoparticles and to enhance turbidity and COD removal.

4. Conclusions

The one-factor-at-a-time method revealed that the two operating parameters, reaction time and catalyst load, had a significant effect on two output parameters, turbidity and COD removal. Furthermore, the optimal ranges for catalyst load and reaction time were established to be 0.3–0.6 g/L and 20–40 min, respectively. These conditions were then applied to the optimization technique. The Central Composite Design (CCD) matrix of 13 experimental runs and 5 center points, with reaction time (20–40 min) and catalyst load (0.3–0.6 g/L) as the 2 input variables, were developed at their low, high, and middle points. RSM was used to model and optimize the input-output variable relationship. The analysis of variance (ANOVA) revealed that all models generated were significant as they possessed a p-value less than 0.05 at a 95% confidence level. Therefore, this signifies good predictability for the optimization of the process. A desirability of 74.7% was achieved at optimal conditions of 0.6 g/L for catalyst load and 40 min for reaction time. The predicted results from RSM were found to be in good agreement with the experimental results. The optimal responses achieved were 32.64% COD removal and 95.17% turbidity removal. The optimal conditions were also applied to the comparative experiment that utilized UV-Vis light instead of UV light. The optimal responses achieved for this were 25.58% COD removal and 66.88% turbidity removal. This is possibly due to TiO₂ not being photoexcited by visible light due to its large energy band gap. Due to the turbidity of the TiO₂, additional post-treatment separation methods are required. Further studies making use of TiO₂ as a catalyst for photocatalytic wastewater treatment should include additional post-treatment filtration steps to recover the TiO₂ to ensure optimal removal of the nanoparticles and to enhance turbidity and COD removal. The use of actual wastewater highlighted the need for better characterization of wastewater for better comparison and explanation of variation in removal efficiencies of different systems.