Abstract

In order to optimize the operating conditions for a combined polyaluminum chloride (PACl) coagulation/flocculation and ultrafiltration process for treating potable water, the main, second order and interaction effects of PACl dose and flocculation retention time (FRT) on permeate turbidity, UV254 and membrane permeability were investigated using a 100 kDa hollow fiber membrane operated in the dead-end mode. A multilevel factorial design was used to determine the relevant ranges of the two factors for optimization. A 22 central composite design (CCD) was then used to develop mathematical correlation models for the optimum operating conditions. The main effect of PACl dose was the most significant factor on all the responses. For permeability, both the main effect of FRT and FRT–PACl dose interactions were found to be insignificant. The optimum PACl dose and FRT for the feed water were 20 mg/L and 14 min, respectively. Corresponding permeate turbidity, UV254 and permeability were 0.15 ± 0.01 NTU, 0.003 ± 0.001 cm−1 and 62.0 ± 9.52 Lm−2 h−1 bar−1, respectively. Experimental validation runs confirmed the reliability of the predicted optimal conditions thus implying that CCD models can be used to predict/optimize the quality and quantity of permeate from hybrid coagulation–ultrafiltration systems for potable water treatment.

INTRODUCTION

Ultrafiltration (UF) is widely applied as an effective technique for removing a diverse array of waterborne pathogens such as protozoa, bacteria and viruses from drinking water (Hill et al. 2007; Mull & Hill 2012). At the appropriate molecular weight cut-off (MWCO), UF membranes can produce bacteria-free drinking water, remove greater than 6 log of viruses, turbidity, a portion of natural organic matter (NOM) and disinfection byproduct precursors from the feed water (Liang et al. 2008; Mimoso et al. 2015). Compared with conventional sand filtration, UF has several advantages which include a higher treatment efficiency, smaller footprint, fewer chemicals demand, less sludge production and easy automation. Furthermore, recent advances in membrane technology have made UF a cost-effective alternative to other filtration techniques (Jeong et al. 2012; Mull & Hill 2012). It is therefore seen by communities as a safer treatment alternative (Guo et al. 2009).

Despite these advantages, membrane fouling still remains an important limitation for this technology (Arhin et al. 2016). Membrane fouling occurs when organic and inorganic matter retained by the membrane accumulate on the membrane surface or within the membrane pores generating an increasing hydraulic resistance. Fouling reduces hydraulic permeability, increases transmembrane pressure and cleaning frequencies, which eventually lead to higher operating costs. Moreover, frequent membrane cleaning operations may lead to membrane deterioration causing reduction in the permeate quality and shorter membrane life span (Chon et al. 2012; Munla et al. 2012; Xu et al. 2014). In order to minimize fouling, pretreatment of feed water using techniques such as coagulation, adsorption, oxidation, bio-filtration, magnetic ion exchange resins and integrated pretreatment is practiced in full-scale and pilot-scale UF plants (Gao et al. 2011). Among the various pretreatment techniques, coagulation is the most dominantly used due to its low cost, ease of operation and ability to significantly improve UF performance (Arhin et al. 2016). It is presumed that during coagulation, colloids are destabilized and clusters form larger flocs which are easily retained by UF membranes (Xiangli et al. 2008). Besides fouling minimization, the use of coagulation pretreatment has an additional beneficial effect since it can improve the removal efficiency of harmful contaminants such as viruses (Fiksdal & Leiknes 2006; Konieczny et al. 2006).

Although coagulation pretreatment is rapidly gaining attention, research evaluating its effect on membrane fouling has been inconsistent. Whereas several studies have reported improved membrane performance with coagulation pretreatment (Dong et al. 2007; Kimura et al. 2014; Lai et al. 2015), others observed decreased or no improvement in membrane performance (Maartens et al. 1999; Howe & Clark 2006). The inconsistencies in coagulation pretreatment have been attributed to several factors such as the specific characteristics of the feed water and properties of NOM, type of membrane material and the module configuration, flow configuration (dead-end or cross-flow), running mode (constant pressure or constant flux), type of coagulant, dose used, whether coagulation was performed with or without flocculation, the retention time used for flocculation and whether there was sedimentation prior to filtration (Arhin et al. 2016). Among these factors, the coagulant dose and the flocculation retention time (FRT) are two factors that have a significant impact on operational costs (Li et al. 2011). These two factors can influence floc formation and the size distribution of flocs, which affect both hydraulic permeability and solute rejection. Moreover, the floc size required in hybrid coagulation-UF systems is not necessarily the same as the floc size required for conventional coagulation because UF membranes are able to retain even small flocs. It is therefore imperative that the coagulant dose and FRT are optimized with regards to membrane performance.

In recent years, response surface methodology (RSM) has become very popular for simultaneous optimization and modelling of chemical processes (Bas & Boyaci 2007; Bezerra et al. 2008). RSM is a collection of statistical and mathematical techniques used to predict and optimize a response or set of responses of interest by adjusting several variables. The main advantage of RSM is that it includes interaction effects among the variables and, thus, depicts the complete effects of the variables on the response unlike the classical one-variable-at-a-time optimization approach where the desired response is obtained by keeping all other variables at a constant level (Prasad et al. 2011). In addition to that, it reduces the number of experimental trials needed to evaluate multiple variables and their interactions, thereby saving time and expenses (Sharma et al. 2009). Yet, most studies optimizing the operating conditions for hybrid coagulation–membrane filtration process have not considered the possibility of using RSM.

Previous studies such as Zularisam et al. (2009) delivered very relevant insights about optimization of alum towards NOM removal and membrane permeability amelioration. Nowadays, there is a growing interest in the use of pre-hydrolyzed coagulants such as polyaluminum chloride (PACl) over alum for coagulation pretreatment. PACl contains highly charged tridecamer cationic species (Al13) which are strongly absorbed on negative colloids (Duan & Gregory 2003). There are several benefits of PACl over alum such as the ability to form more robust flocs, lower aluminum residuals in treated water, reduced sludge volumes and improved performance under low temperature regimes (Arhin et al. 2016). Aside from that, in addition to NOM removal, it is also important to consider the influence of coagulation on the nexus between turbidity and reversible fouling as turbidity spikes affect plant operational and maintenance regimes such as backflush flux and frequency, chemical cleaning protocol and transmembrane pressure which in tend influence fouling rates and productivity (Amy 2008; Hatt et al. 2011). Moreover, some studies have demonstrated fouling suppression at coagulant doses not significantly influencing NOM removal, thus suggesting flux amelioration is not contingent upon NOM removal (Choi & Dempsey 2004; Konieczny et al. 2009).

In this study, the optimum operating conditions for a hybrid PACl coagulation-UF system for potable water treatment were determined using RSM. The focus of this study was to establish the optimum PACl dose and FRT for enhancing both organic and inorganic foulants removal. To that end, the independent variables considered were PACl dose and FRT, and the process output was assessed in terms of UV254 removal: used as a surrogate for NOM, especially for humic acids (Cho et al. 2006), turbidity removal: used as an indicator for particulate matter (Kim et al. 2007) and membrane permeability. Preliminary batch experiments (jar tests) were performed to test the relevance of the variables on the feed water and to determine their ranges for optimization. Thereafter by means of RSM, the optimum PACl dose and FRT were determined. This study provides insights into quantitative identification and simultaneous optimization of significant variables for hybrid coagulation-UF systems.

MATERIALS AND METHODS

Raw water sampling and samples characterization

Surface water samples from Lake Victoria were collected at the raw water inlet to Ggaba II Water Treatment Works, Kampala, Uganda from November 2015 to April 2016. Daily sampling was done between 8:00 and 9:00 am. Analytical tests were conducted to characterize the raw water based on temperature, pH, electrical conductivity, turbidity, color, and UV254. Temperature was determined with an Extech MO295 IR thermometer (Extech® instruments, USA). pH and conductivity were determined using Hach Sension + MM374 multi parameter meter (Hach, USA). A turbidimeter (Hach 2100Q, USA) was used for turbidity measurements and a UV spectrophotometer (Hach DR1900, USA) was used for color determination at 455 nm. Another UV spectrophotometer (photoLab 6600 UV-VIS; WTW, Germany) was used for UV254. Following standard procedure, samples for UV254 analysis were filtered with 0.45 μm sterile filter membranes (MCNE-447-100; Labbox, France) prior to analysis (USEPA 2003). Samples which could not be analyzed on the day of sampling were kept at a temperature of +4 °C (Gamage & Chellam 2011). No sample was kept for more than 3 days without performing analysis. Table 1 summarizes the characteristics of the feed water used for the study. Each parameter was determined at least in triplicate.

Table 1

Raw water quality of Lake Victoria

Parameter Mean Standard deviation 
Temperature (°C) 24.7 1.4 
pH 7.2 0.3 
Conductivity (μS/cm) 120.6 2.1 
Turbidity (NTU) 15.4 2.5 
Color (PtCo) 177.0 17.0 
UV254 (cm−10.030  0.010 
Parameter Mean Standard deviation 
Temperature (°C) 24.7 1.4 
pH 7.2 0.3 
Conductivity (μS/cm) 120.6 2.1 
Turbidity (NTU) 15.4 2.5 
Color (PtCo) 177.0 17.0 
UV254 (cm−10.030  0.010 

Preliminary batch experiments

Preliminary experiments were done to verify the statistical relevance of the two independent variable (PACl dose and FRT) and to determine their experimental ranges for optimization (Choi & Dempsey 2004; Cho et al. 2006; Bas & Boyaci 2007). A multilevel factorial design consisting of two factors, 24 base runs and three replicates was used to determine the statistical significance of PACl dose and FRT on the quality of water from Lake Victoria. The supernatant turbidity, UV254 and color were used to denote the quality characteristics. A six paddle jar test apparatus (Stuart Scientific Flocculator; SW1, UK) was used to study the change in particle size during coagulation. Tests were executed following standard protocol (ASTM 2013). Commercial PACl in liquid form (Zetafloc 553 l) was purchased from Abby Laboratories, South Africa. The stock solution was prepared by dosing one mL of the coagulant in 100 mL of distilled water. Coagulation/flocculation was done in 1,000 mL Griffin beakers. The multiple stirrer was started at a speed of 120 rpm and coagulant doses corresponding to 10, 12, 14, 16, 18, and 20 mg/L were added without pH adjustment: previous studies have reported measureable fouling abatement within the pH range measured for the raw water used (Hatt et al. 2011). Rapid mixing was done for a minute after which the speed was reduced to 30 rpm. The experiment was performed under FRTs of 5, 10, 15, and 20 min. After slow mixing, the paddles were withdrawn and samples were settled for 15 min. Supernatant samples were then siphoned mid-way through the beakers for turbidity, color and UV254 analysis.

Membrane filtration

Commercially available hollow tube UF membranes were used. The membrane specifications are provided in Table 2. The experimental setup is schematically illustrated in Figure 1. As shown in Figure 1, the set-up consists of coagulation/flocculation process followed by UF without prior sedimentation (Bergamasco et al. 2011). The experiment was conducted in the constant pressure mode so that any decline in permeate flux due to membrane fouling could be monitored effectively (Meng et al. 2006). A constant transmembrane pressure of 30 kPa was applied to extract permeate across the membrane in the dead-end mode. A speed variable peristaltic pump (AT300S + YT25 pump head; Dalian Tiancheng Machine Co. Ltd, China) was used to pump the coagulated water through the UF membrane at a constant pump speed. The pump served a dual function as feed and backwash pump.

Table 2

Membrane specification

Parameter Membrane module 
Membrane type Hollow fiber 
Supplier Frontec Environmental Co., Ltd, China 
Module configuration Inside-out 
MWCO 100 kDa 
Module length 0.533 m 
Membrane area 1.9 m2 
Parameter Membrane module 
Membrane type Hollow fiber 
Supplier Frontec Environmental Co., Ltd, China 
Module configuration Inside-out 
MWCO 100 kDa 
Module length 0.533 m 
Membrane area 1.9 m2 
Figure 1

Schematic representation of the hybrid coagulation-UF process.

Figure 1

Schematic representation of the hybrid coagulation-UF process.

The membrane permeate flow was determined by weighing permeate on a top-loading balance at timed intervals (Lai et al. 2015). The permeate flux was then obtained by dividing the permeate flow by the membrane surface area as shown in Equation (1):  
formula
(1)
where Q is the permeate flow (L/h) and A is the membrane area (m2). The membrane permeability – after 60 min of continuous filtration – was subsequently determined by dividing the flux by the transmembrane pressure as shown in Equation (2):  
formula
(2)
where TMP is the transmembrane pressure (Pa).

Experimental design for RSM

After the experimental ranges for the independent variables were obtained, RSM was used to determine the optimum factor settings for pre-coagulation. A 22 rotatable central composite design (CCD) consisting of factor combinations at two levels, four star points and five replicates at the center point was developed. The coagulant dose (coded as X1) and FRT (coded as X2) were set as the independent variables while the permeate turbidity (Coded as Y1), UV254 absorbance (coded as Y2) and the membrane permeability (coded as Y3) were set as the response variables. Randomization of the experimental runs was done to limit the effects of unexpected variability in the response. Table 3 shows the range and levels of the two independent variables for the response surface design.

Table 3

Independent variables and factor levels for response surface design

  Coded levels
 
  −1.4 −1.0 0.0 + 1.0 + 1.4 
Factors Units Natural levels 
Dose (X1mg/L 7.0 10.0 15.0 20.0 22.0 
FRT (X2Min 3.3 5.0 10.0 15.0 17.0 
  Coded levels
 
  −1.4 −1.0 0.0 + 1.0 + 1.4 
Factors Units Natural levels 
Dose (X1mg/L 7.0 10.0 15.0 20.0 22.0 
FRT (X2Min 3.3 5.0 10.0 15.0 17.0 
The experimental data from RSM was fitted into the polynomial model indicated in Equation (3):  
formula
(3)
where Y is the response variable; X1, X2,…, Xk are the independent variables affecting the response Y; β0, βi (i = 1, 2,…, k), βii (i = 1, 2,…, k) and βij (i = 1, 2,…, k; j = 1, 2,…, k) are the regression coefficients for intercept, linear, quadratic and interaction terms respectively; and k is the number of variables (Bas & Boyaci 2007; Silva et al. 2007).
After developing mathematical models for the response variables, the Derringer function (Murphy et al. 2005; Bezerra et al. 2008) was used to obtain a desirability function for each individual response (di; 0 ≤ di ≤ 1), after which the overall desirability function (D), was computed as a weighted average of the individual desirability (di) according to Equation (4):  
formula
(4)
where m is the number of responses.
The goal of the coagulation-UF process is to obtain a low turbidity and low UV254 (organics) in the permeate while maintaining a relatively high permeability across the UF membrane. Therefore, the desirability function for turbidity and UV254 were both the ‘smaller-the-better’ responses, which criterion is expressed in Equation (5) whereas the desirability function for permeability was the ‘larger-the-better’ response, which transformation is described in Equation (6):  
formula
(5)
 
formula
(6)
where U is the upper acceptable value to the response, L is the lower acceptable value to the response, T is the target value, y is the response and t is the weight (Bezerra et al. 2008).

After the optimum conditions were obtained from the desirability function, four experimental replicates were performed at the optimal settings (Silva et al. 2007). The experimental and predicted values were subsequently compared to validate the response surface optimization.

Data analysis and statistical testing

Minitab 17.3.1 (Minitab®, USA) was used for statistical testing. All statistical tests were conducted at 95% confidence interval (CI). The statistical significance of mathematical correlation models was verified by Fisher's statistical test (F-test) for analysis of variance (ANOVA). Also, the coefficient of determination, R2 was used as an indicator of the adequacy of the fitted polynomial model.

RESULTS AND DISCUSSION

Preliminary jar test results

The statistical significance of PACl dose and FRT on turbidity, UV254 and color of the feed water was investigated and their relevant experimental ranges to consider for optimization were determined. As can be inferred from Table 4, both PACl dose and FRT were found to be significant factors affecting turbidity, UV254 and color. With the exception of UV254, a p-value of 0.000 was computed for both PACl dose and FRT on all the responses indicating that these factors could highly influence the water quality. A similar observation was reported by Li et al. (2011) who claimed that among the factors affecting the hybrid coagulation – UF process, coagulant dose and retention time are two crucial factors that can impact on floc formation and size distribution as well as permeability of UF membranes. Thus, a remarkable improvement on the water quality coupled with improved membrane permeability could be attained if those two factors could be adjusted correctly.

Table 4

Statistical results for preliminary jar test

Response Source Sums of squares DF Mean squares F-value p-value 
Turbidity PACl dose 18.500 3.7001 8.88 0.000 
FRT 13.273 4.4244 10.61 0.000 
UV254 PACl dose 0.000809 0.000162 2.60 0.037 
FRT 0.000712 0.000237 3.82 0.016 
Color PACl dose 1421.9 284.39 18.56 0.000 
FRT 1,832.8 610.93 27.87 0.000 
Response Source Sums of squares DF Mean squares F-value p-value 
Turbidity PACl dose 18.500 3.7001 8.88 0.000 
FRT 13.273 4.4244 10.61 0.000 
UV254 PACl dose 0.000809 0.000162 2.60 0.037 
FRT 0.000712 0.000237 3.82 0.016 
Color PACl dose 1421.9 284.39 18.56 0.000 
FRT 1,832.8 610.93 27.87 0.000 

Figure 2 shows the effect of varying PACl dose and FRT on turbidity, UV254 and color. As depicted in Figure 2(a), turbidity removal did not improve beyond 15 min. Even though, at 20 min FRT, a slight improvement was observed from 12 to 14 mg/L of the coagulant, a much turbid supernatant was attained afterwards. For UV254 and color, increasing the FRT to 20 min produced a much poorer supernatant quality. These results were consistent with the findings of Cho et al. (2006) who observed that at retention times larger than 20 min, there was no improvement in the removal of UV254. Similarly, Li et al. (2011) noted that beyond 10 min retention time, no significant improvement in the removal of dissolved organics could be achieved with coagulant doses ranging from 15 to 25 mg/L. It appears that the removal of organics and particulates mostly occurs at the early stages of flocculation and depending on the specific characteristics of the raw water and the parameter being studied, there is a certain critical FRT beyond which no significant improvement in water quality can be attained. In this study, that critical FRT was observed at 15 mins. Consequently, 20 min FRT was not considered during the optimization process as a high retention time implicates higher mixing volumes as well as a higher energy consumption and yet no improvement in water quality would be attained. Therefore PACl doses ranging from 10 to 20 mg/L and FRT ranging from 5 to 15 min were used for RSM optimization.

Figure 2

Effect of varying PACl dose and FRT on Turbidity (a), UV254 (b) and color (c) removal.

Figure 2

Effect of varying PACl dose and FRT on Turbidity (a), UV254 (b) and color (c) removal.

Optimization of the hybrid coagulation-UF system

Table 5 shows the layout and results of the 22 CCD. Irrespective of the factor settings used in this study, the color of permeate was zero (100% removal of color forming compounds). Hence, color was deemed an inappropriate response for assessing the effect of varying PACl dose and FRT on the hybrid coagulation-UF system. Three quadratic models representing the permeate turbidity, the presence of UV254 absorbing organics and the permeability across the membrane were therefore obtained from the CCD and their statistical significance were examined by ANOVA. Fit summary output analysis revealed that the models were statistically significant (p-value <0.05) to describe the permeate turbidity, UV254 and the membrane permeability.

Table 5

Two-factorial CCD layout and experimental results

Standard ordera Run orderb Blocks Factor variables
 
Response variablesc
 
X1 X2 Y1 (NTU) Y2 (cm−1Y3 (Lm−2 h−1bar−1
0.0 1.4 0.22 0.006 56.5 
−1.4 0.0 0.25 0.007 64.6 
−1.0 1.0 0.22 0.008 68.0 
0.0 0.0 0.16 0.002 28.8 
12 0.0 0.0 0.16 0.002 34.6 
1.0 1.0 0.16 0.004 68.4 
1.0 −1.0 0.18 0.001 56.6 
0.0 −1.4 0.25 0.002 59.0 
11 0.0 0.0 0.17 0.002 46.2 
10 −1.0 −1.0 0.29 0.003 45.4 
11 10 0.0 0.0 0.17 0.003 46.1 
12 13 0.0 0.0 0.16 0.002 35.8 
13 1.4 0.0 0.13 0.002 68.7 
Standard ordera Run orderb Blocks Factor variables
 
Response variablesc
 
X1 X2 Y1 (NTU) Y2 (cm−1Y3 (Lm−2 h−1bar−1
0.0 1.4 0.22 0.006 56.5 
−1.4 0.0 0.25 0.007 64.6 
−1.0 1.0 0.22 0.008 68.0 
0.0 0.0 0.16 0.002 28.8 
12 0.0 0.0 0.16 0.002 34.6 
1.0 1.0 0.16 0.004 68.4 
1.0 −1.0 0.18 0.001 56.6 
0.0 −1.4 0.25 0.002 59.0 
11 0.0 0.0 0.17 0.002 46.2 
10 −1.0 −1.0 0.29 0.003 45.4 
11 10 0.0 0.0 0.17 0.003 46.1 
12 13 0.0 0.0 0.16 0.002 35.8 
13 1.4 0.0 0.13 0.002 68.7 

aNot randomized.

bRandomized.

cMean of three readings.

As depicted in Table 6, the p-value for the mathematical model for turbidity implied that the model was highly significant. In addition to that, the multiple regression coefficient R2 of the model was 0.9850, suggesting that 98.50% of the variability in the response could be well explained by the model. Thus, a good agreement existed between the experimental and the predicted values as only 1.50% of the variability was poorly described by the model. Moreover, the value for Lack-of-Fit was insignificant denoting that the model was reasonable fit. The normal probability and response surface plots for turbidity are displayed in Figure 3. The analysis of the CCD is based on the assumption that the observations made are independently and normally distributed. As shown by the normal probability plot, the data points were fairly close to the straight line thus depicting that the experiment came from a normally distributed population. It can also be seen from the surface plot that the permeate turbidity decreased with increasing PACl dose. The minimum turbidity (<0.15 NTU) was observed with PACl dose >20 mg/L and FRT ranging from 6–13.5 min (Figure S1, Supplement, available with the online version of this paper).

Table 6

Statistical analysis of the results of the two-factorial CCD

Response Factors SS DF MS F-value p-value 
Turbidity Model 0.026505 0.005301 92.10 0.000 
 X1 0.014425 0.014425 250.61 0.000 
 X2 0.002192 0.002192 38.08 0.000 
 X1*X1 0.001176 0.001176 20.42 0.003 
 X2*X2 0.008767 0.008767 152.31 0.000 
 X1*X2 0.000625 0.000625 10.86 0.013 
 Lack-of-Fit 0.000283 0.000094 3.14 0.149 
 R2 = 0.9850      
 R2-Adj =0.9743      
UV254 Model 0.000057 0.000011 45.74 0.000 
 X1 0.000021 0.000021 85.20 0.000 
 X2 0.000023 0.000023 93.01 0.000 
 X1*X1 0.000008 0.000008 32.82 0.001 
 X2*X2 0.000005 0.000005 19.47 0.003 
 X1*X2 0.000001 0.000001 3.99 0.085 
 Lack-of-Fit 0.000001 0.000000 1.59 0.330 
 R2 = 0.9706      
 R2-Adj = 0.9496      
Permeability Model 45881.3 9176.3 5.95 0.018 
 X1 949.0 949.0 0.61 0.030 
 X2 2992.8 2992.8 1.94 0.168 
 X1*X1 31794.4 31794.4 20.60 0.003 
 X2*X2 14285.2 14285.2 9.26 0.018 
 X1*X2 718.1 718.1 0.47 0.517 
 Lack-of-Fit 4954.3 1651.4 1.13 0.439 
 R2 = 0.8100      
 R2-Adj =0.6742      
Response Factors SS DF MS F-value p-value 
Turbidity Model 0.026505 0.005301 92.10 0.000 
 X1 0.014425 0.014425 250.61 0.000 
 X2 0.002192 0.002192 38.08 0.000 
 X1*X1 0.001176 0.001176 20.42 0.003 
 X2*X2 0.008767 0.008767 152.31 0.000 
 X1*X2 0.000625 0.000625 10.86 0.013 
 Lack-of-Fit 0.000283 0.000094 3.14 0.149 
 R2 = 0.9850      
 R2-Adj =0.9743      
UV254 Model 0.000057 0.000011 45.74 0.000 
 X1 0.000021 0.000021 85.20 0.000 
 X2 0.000023 0.000023 93.01 0.000 
 X1*X1 0.000008 0.000008 32.82 0.001 
 X2*X2 0.000005 0.000005 19.47 0.003 
 X1*X2 0.000001 0.000001 3.99 0.085 
 Lack-of-Fit 0.000001 0.000000 1.59 0.330 
 R2 = 0.9706      
 R2-Adj = 0.9496      
Permeability Model 45881.3 9176.3 5.95 0.018 
 X1 949.0 949.0 0.61 0.030 
 X2 2992.8 2992.8 1.94 0.168 
 X1*X1 31794.4 31794.4 20.60 0.003 
 X2*X2 14285.2 14285.2 9.26 0.018 
 X1*X2 718.1 718.1 0.47 0.517 
 Lack-of-Fit 4954.3 1651.4 1.13 0.439 
 R2 = 0.8100      
 R2-Adj =0.6742      
Figure 3

Normal probability plot (a) and response surface plot (b) for turbidity.

Figure 3

Normal probability plot (a) and response surface plot (b) for turbidity.

As shown in Table 6, the main effect of PACl dose (X1), the main effect of FRT (X2), the second order effect of PACl dose (X1*X1), the second order effect of FRT (X2*X2) and the interaction of PACl dose and FRT (X1*X2) were all found to be the significant factors affecting the permeate turbidity. The regression model for turbidity (Y1) is given by Equation (7):  
formula
(7)
Apart from turbidity, UV254 is another important response variable for this study as it was used to connote the efficiency of pre-coagulation in removing organics from the feed water. As depicted in Table 6, the model for UV254 was found to be very significant. Coupled with that, the R2 showed that 97.06% of the variability in the response could be well explained by the model. Furthermore, the model was found to be satisfactory as the value for Lack-of-Fit was statistically insignificant. The normal probability and surface plot for UV254 are shown in Figure 4. As shown in Figure 4, UV254 removal increased with increasing PACl dose and decreasing FRT. The lowest UV254 (<0.002 cm−1) could be achieved with PACl dose >16 mg/L and at a retention time <11 min (Figure S2, Supplement, available online).
Figure 4

Normal probability plot (a) and response surface plot (b) for UV254 absorbance.

Figure 4

Normal probability plot (a) and response surface plot (b) for UV254 absorbance.

The significant model terms for UV254 were the main effect of PACl dose (X1), the second order effect of PACl dose (X1*X1), the main effect of FRT (X2) and the second other effect of FRT (X2*X2). The two level interactions of PACl dose and FRT was statistically insignificant (p-value = 0.085) and was therefore eradicated using backward elimination procedures to improve the model (Table S1, Supplement, available online). Ultimately, the regression model for UV254 absorbance (Y2) was reduced to the form shown in Equation (8):  
formula
(8)
Permeability as a third response variable was used to denote the fouling potentials of the pre-coagulated water. Less fouling is accompanied by high hydraulic permeability while the converse is true for exacerbated fouling (Bae & Tak 2005). The empirical model for permeability was also found to be statistically significant (Table 6). An R2 of 0.8100 was computed for the regression indicating that there is a strong correlation. To cap it all, the Lack-of-Fit test did not show inadequacy of the model implying that the model could adequately fit the experimental results. The normal probability and surface plots for permeability are also shown in Figure 5. As depicted in response surface plot, permeability decreased around the center of the surface plot. The minimum permeability (28.8 Lm−2 h−1 bar−1) was centered in the vicinity of 17 mg/L PACl and FRT of 9 min with increasing values along both sides of the point (Figure S3, Supplement, available online).
Figure 5

Normal probability plot (a) and response surface plot (b) for permeability.

Figure 5

Normal probability plot (a) and response surface plot (b) for permeability.

It was also revealed that within the context of this study, the main effect of PACl dose (X1), the second order effect of PACl dose (X1*X1) and the second order effect of FRT (X2*X2) were the significant factors affecting the permeability (Table 6). Therefore to improve the model, the backwards elimination procedure was once again employed to eradicate the insignificant factors (Table S1, Supplement). The final empirical model for permeability is shown in Equation (9):  
formula
(9)
The membrane permeability seemed to be dependent on a complex interaction between the coagulant dose and FRT. Coagulant doses and FRTs that formed smaller flocs had low permeability. On the other hand, coagulant doses and FRTs that formed bigger flocs contributed to high membrane permeability. This corresponds to the cake layer theory according to the Carman-Kozeny model (Equation (10)), which implies that the cake layer resistance is inversely proportional to the square of the particle diameter:  
formula
(10)
where Kc is the specific cake resistance, Cc is solid concentration or solidosity of the cake in volume percentage and dp is the particle diameter.

This was also observed in previous studies demonstrating that a cake layer formed by smaller particles is much compact compared to that formed by large ones, which imply that smaller particles have a higher specific cake resistance (Lin et al. 2011; Chomiak et al. 2014). In addition to that, coagulant doses and FRTs that formed bigger flocs also contributed to higher removal of UV254 organics. Because humic NOM has more electron rich sites (carboxylic functional groups) for bulk NOM agglomeration through charge neutralization with cationic aluminum hydrolysis, at the appropriate coagulation conditions, humic acid molecules are more readily absorbed to the surface of positively charged Al species (Choi et al. 2013). Better NOM agglomeration would therefore imply the formation of bigger flocs which were easily retained by the UF membrane and this explains why higher UV254 organics removal was observed with bigger flocs. It must also be noted that coagulant doses and FRTs that formed very tiny flocs had low permeate quality due to the permeation of smaller particles through the UF membrane.

Determination and experimental validation of the optimum results

For any given condition, a tradeoff was found between turbidity, UV254 and the permeate flux. High turbidity and UV254 rejections were sometimes accompanied by low permeability and vice versa. This phenomenon is very evident in standard orders 4, 5 and 12 (Table 5) where turbidity of 0.16 NTU and UV254 absorbance of 0.002 cm−1 corresponded with very low permeability (28.8, 34.6 and 35.8 Lm−2 h−1 bar−1 for standard orders 4, 5 and 12, respectively). These results were consistent with previous studies by Zularisam et al. (2009) who observed a contradictory relation between the membrane permeability and NOM rejection in that an increase in NOM rejection through a single separation process sometimes led to a decrease in membrane permeability and vice versa. Hence, to simultaneously optimize the responses, a multi-criteria approach that finds the optimal compromises between turbidity, UV254 and membrane permeability was required to predict the optimum factor combination. Using the desirability function approach (Murphy et al. 2005), the composite desirability of the three responses was established (Figure S4, Supplement, available online). At this composite desirability, the predicted optimal coagulant dose and FRT were 20.07 (≈20) mg/L and 13.64 (≈14) min, respectively. Corresponding turbidity, UV254 and permeability at 95% CI were 0.15 ± 0.01 NTU, 0.0033 ± 0.0007 cm−1 and 62.0 ± 9.52 Lm−2 h1 bar−1. In order to confirm the prediction accuracy of the empirical models, the adequacy of the models were validated through experiments. Experimental confirmation runs under the optimized conditions confirmed the validity of the models as the results of the experiment closely agreed with the models’ predictions (Table 7). In addition to that there was no significant difference (p > 0.05) between the predicted and observed values. Thus the models can be reliably used to predict the quality and quantity of permeate from the hybrid coagulation-UF membrane process of feed water from Lake Victoria at the experimental scale.

Table 7

Comparison between predicted and observed values for turbidity, UV254 absorbance and membrane permeability

Optimized conditiona 
 Dose (mg/L) 20.07 (≈20) 
 FRT (min) 13.643 (≈14) 
Predicted valuesb 
 Turbidity (NTU) 0.15 ± 0.01 
 UV254 absorbance (cm−10.0033 ± 0.0007 
 Permeability (Lm−2 h−1bar−162.0 ± 9.52 
Observed valuesc 
 Turbidity (NTU) 0.15 ± 0.01 
 UV254 absorbance (cm−10.0030 ± 0.0010 
 Permeability (Lm−2 h−1 bar−159.0 ± 2.1 
Optimized conditiona 
 Dose (mg/L) 20.07 (≈20) 
 FRT (min) 13.643 (≈14) 
Predicted valuesb 
 Turbidity (NTU) 0.15 ± 0.01 
 UV254 absorbance (cm−10.0033 ± 0.0007 
 Permeability (Lm−2 h−1bar−162.0 ± 9.52 
Observed valuesc 
 Turbidity (NTU) 0.15 ± 0.01 
 UV254 absorbance (cm−10.0030 ± 0.0010 
 Permeability (Lm−2 h−1 bar−159.0 ± 2.1 

aObtained from desirability function.

bMean ± CI.

cMean ± CI (n = 4).

It should be kept in mind that the present model is focused on permeability, turbidity and UV254 removal only. For practical implementation of the coagulation-UF process, the end points which need to be optimized actually are water quality (turbidity, organics, pathogen count) and total annual costs. Membrane flux influences investment costs and capital costs (depreciation). However, total costs are furthermore related to the consumption of chemicals, including coagulant and cleaning chemicals, and other factors such as labor and maintenance. These factors depend highly on locally and dependent variables such as interest rate and wages. Thus, optimizing operation costs is far more complex and exceeds the framework of this study. However, in principle a similar approach can be used whereby the desirability is extended with functions describing capital costs and the costs for chemical usage and labor (cleaning operations) in order to optimize the overall feasibility of the coagulation-UF process.

CONCLUSIONS

The optimum operating conditions for a hybrid PACl coagulation-UF system for treating potable water from Lake Victoria were obtained using RSM. The focus of the study was to establish the optimum PACl dose and FRT for enhancing both organic and inorganic matter removal while simultaneously optimizing membrane permeability. Using RSM models, the optimum PACl dose and FRT for the feed water were 20.07 (≈20) mg/L and 13.64 (≈14) min, respectively. At this factor combination, turbidity, UV254 and permeability of 0.15 ± 0.01 NTU, 0.0030 ± 0.0010 cm−1 and 62.0 ± 9.52 Lm−2 h−1 bar−1 were respectively predicted. Experimental validations confirmed the reliability of these predicted optimum settings, indicating that RSM models can be used to predict optimum operating conditions for hybrid coagulation-UF process of the feed water at the bench scale.

The effectiveness of the hybrid system appears to be influenced by a complex interaction between the coagulant dose and FRT. Coagulant doses and FRTs that formed smaller flocs offered high hydraulic resistance which led to low membrane permeability whereas agglomeration of particles into bigger flocs contributed to high membrane permeability. Also, smaller flocs led to low turbidity and UV254 rejection while the converse was true for bigger flocs. However, a tradeoff was found between the permeate turbidity, UV254 and the permeability in that increase in one sometimes led to a decrease in the other. Consequently, a multi-criteria approach that finds the optimal compromises between turbidity, UV254 and membrane permeability was used to simultaneously optimize the responses. This demonstrates that to maximize production, coagulation conditions should be set in order to achieve the optimum compromise between floc size, solute rejection and membrane permeability. The results presented here provide insights into simultaneous optimization of relevant operating factors in pilot or full scale coagulation-UF plants for drinking water treatment.

ACKNOWLEDGEMENTS

This study was supported by the Mobility to Enhance Training of Engineering Graduates in Africa (METEGA) Project, the Regional Universities Forum for Capacity Building in Agriculture (RUFORUM) [grant number: RU 2015 FAPA 063] and the Swiss Federal Institute of Aquatic Science and Technology (Eawag) through the Eawag Partnership Program for Developing Countries (EPP). We thank National Water and Sewerage Corporation, Uganda for granting us laboratory space. The sponsors of the study had no role in study design; in the collection, analysis and interpretation of data; in the writing of the report; or in the decision to submit the article for publication.

REFERENCES

REFERENCES
Arhin
,
S. G.
,
Banadda
,
N.
,
Komakech
,
A. J.
,
Kabenge
,
I.
&
Wanyama
,
J.
2016
Membrane fouling control in low pressure membranes : a review on pretreatment techniques for fouling abatement
.
Environmental Engineering Research
21
(
2
),
109
120
.
ASTM International D2035
2013
Standard Practice for Coagulation-Flocculation Jar Test of Water
,
ASTM International
,
West Conshohocken, PA
,
USA
.
Bas
,
D.
&
Boyaci
,
I. H.
2007
Modeling and optimization I : usability of response surface methodology
.
Journal of Food Engineering
78
(
3
),
836
845
.
Bergamasco
,
R.
,
Konradt-Moraes
,
L. C.
,
Vieira
,
M. F.
,
Fagundes-Klen
,
M. R.
&
Vieira
,
A. M. S.
2011
Performance of a coagulation-ultrafiltration hybrid process for water supply treatment
.
Chemical Engineering Journal
166
(
2
),
483
489
.
Bezerra
,
M. A.
,
Santelli
,
R. E.
,
Oliveira
,
E. P.
,
Villar
,
L. S.
&
Escaleira
,
L. A.
2008
Response surface methodology (RSM) as a tool for optimization in analytical chemistry
.
Talanta
76
(
5
),
965
977
.
Choi
,
K. Y.
&
Dempsey
,
B. A.
2004
In-line coagulation with low-pressure membrane filtration
.
Water Research
38
(
19
),
4271
4281
.
Chomiak
,
A.
,
Mimoso
,
J.
,
Koetzsch
,
S.
,
Sinnet
,
B.
,
Pronk
,
W.
,
Derlon
,
N.
&
Morgenroth
,
E.
2014
Biofilm formation and permeate quality improvement in gravity driven membrane ultrafiltration
.
Water Science & Technology: Water Supply
14
(
2
),
274
282
.
Dong
,
B. Z.
,
Chen
,
Y.
,
Gao
,
N. Y.
&
Fan
,
J. C.
2007
Effect of coagulation pretreatment on the fouling of ultrafiltration membrane
.
Journal of Environmental Science
19
(
3
),
278
283
.
Duan
,
J.
&
Gregory
,
J.
2003
Coagulation by hydrolysing metal salts
.
Advances in Colloids & Interface Science
100
,
475
502
.
Gamage
,
N. P.
&
Chellam
,
S.
2011
Aluminum electrocoagulation pretreatment reduces fouling during surface water microfiltration
.
Journal of Membrane Science
379
(
1
),
97
105
.
Gao
,
W.
,
Liang
,
H.
,
Ma
,
J.
,
Han
,
M.
,
Chen
,
Z.
,
Han
,
Z.
&
Li
,
G.
2011
Membrane fouling control in ultrafiltration technology for drinking water production: a review
.
Desalination
272
(
1
),
1
8
.
Guo
,
X.
,
Zhang
,
Z.
,
Fang
,
L.
&
Su
,
L.
2009
Study on ultrafiltration for surface water by a polyvinylchloride hollow fiber membrane
.
Desalination
238
(
1–3
),
183
191
.
Hatt
,
J. W.
,
Germain
,
E.
&
Judd
,
S. J.
2011
Water treatment using hybrid method of coagulation and low-pressure membrane filtratio
.
Water Research
45
(
1
),
6471
6478
.
Hill
,
V. R.
,
Kahler
,
A. M.
,
Jothikumar
,
N.
,
Johnson
,
T. B.
,
Hahn
,
D.
&
Cromeans
,
T. L.
2007
Multistate evaluation of an ultrafiltration-based procedure for simultaneous recovery of enteric microbes in 100-liter tap water samples
.
Applied & Environmental Microbiology
73
(
13
),
4218
4225
.
Howe
,
K. J.
&
Clark
,
M. M.
2006
Effect of coagulation pretreatment on membrane filtration performance
.
Journal of American Water Works Association
98
(
4
),
133
146
.
Jeong
,
S.
,
Choi
,
Y. J.
,
Nguyen
,
T. V.
,
Vigneswaran
,
S.
&
Hwang
,
T. M.
2012
Submerged membrane hybrid systems as pretreatment in seawater reverse osmosis (SWRO): optimisation and fouling mechanism determination
.
Journal of Membrane Science
411
,
173
181
.
Konieczny
,
K.
,
Bodzek
,
M.
&
Rajca
,
M.
2006
A coagulation-MF system for water treatment using ceramic membranes
.
Desalination
198
(
1–3
),
92
101
.
Konieczny
,
K.
,
Rajca
,
M.
,
Bodzek
,
M.
&
Kwiecinska
,
A.
2009
Water treatment using hybrid method of coagulation and low-pressure membrane filtration
.
Environmental Protection Engineering
35
(
1
),
5
22
.
Li
,
M.
,
Wu
,
G.
,
Guan
,
Y.
&
Zhang
,
X.
2011
Treatment of river water by a hybrid coagulation and ceramic membrane process
.
Desalination
280
(
1–3
),
114
119
.
Liang
,
H.
,
Gong
,
W.
,
Chen
,
J.
&
Li
,
G.
2008
Cleaning of fouled ultrafiltration (UF) membrane by algae during reservoir water treatment
.
Desalination
220
(
1–3
),
267
272
.
Lin
,
H. J.
,
Gao
,
W. J.
,
Leung
,
K. T.
,
Liao
,
B. Q.
&
Lin
,
H. J.
2011
Characteristics of different fractions of microbial flocs and their role in membrane fouling
.
Water Science & Technology
63
(
2
),
262
269
.
Maartens
,
A.
,
Swart
,
P.
&
Jacobs
,
E. P.
1999
Feed-water pretreatment: methods to reduce membrane fouling by natural organic matter
.
Journal of Membrane Science
163
(
1
),
51
62
.
Meng
,
F.
,
Zhang
,
H.
,
Yang
,
F.
,
Li
,
Y.
,
Xiao
,
J.
&
Zhang
,
X.
2006
Effect of filamentous bacteria on membrane fouling in submerged membrane bioreactor
.
Journal of Membrane Science
272
(
1
),
161
168
.
Mimoso
,
J.
,
Pronk
,
W.
,
Morgenroth
,
E.
&
Hammes
,
F.
2015
Bacterial growth in batch-operated membrane filtration systems for drinking water treatment
.
Separation & Purification Technology
156
,
165
174
.
Munla
,
L.
,
Peldszus
,
S.
&
Huck
,
P. M.
2012
Reversible and irreversible fouling of ultrafiltration ceramic membranes by model solutions
.
Journal of American Water Works Association
104
(
10
),
540
554
.
Murphy
,
T. E.
,
Tsui
,
K.
&
Allen
,
J. K.
2005
A review of robust design methods for multiple responses
.
Research in Engineering Design
16
(
4
),
118
132
.
Prasad
,
K. N.
,
Abdulrahman
,
F.
,
Yang
,
B.
,
Weng
,
K.
,
Nagasundara
,
R.
,
Azlan
,
A.
&
Ismail
,
A.
2011
Response surface optimisation for the extraction of phenolic compounds and antioxidant capacities of underutilised Mangifera pajang Kosterm peels
.
Food Chemistry
128
(
4
),
1121
1127
.
Silva
,
E. M.
,
Rogez
,
H.
&
Larondelle
,
Y.
2007
Optimization of extraction of phenolics from Inga edulis leaves using response surface methodology
.
Separation & Purification Technology
55
(
3
),
381
387
.
USEPA Method 415.3
2003
Determination of Total Organic Carbon and Specific UV Absorbance at 254 nm in Source Water and Drinking Water
,
US Environmental Protection Agency
,
Cincinnati, OH
,
USA
.
Xiangli
,
Q.
,
Zhenjia
,
Z.
,
Nongcun
,
W.
,
Wee
,
V.
,
Low
,
M.
,
Loh
,
C. S.
&
Teck Hing
,
N.
2008
Coagulation pretreatment for a large-scale ultrafiltration process treating water from the Taihu River
.
Desalination
230
(
1–3
),
305
313
.
Zularisam
,
A. W.
,
Ismail
,
A. F.
,
Salim
,
M. R.
,
Sakinah
,
M.
&
Matsuura
,
T.
2009
Application of coagulation – ultrafiltration hybrid process for drinking water treatment : optimization of operating conditions using experimental design
.
Separation & Purification Technology
65
(
2
),
193
210
.

Supplementary data