Abstract

Many studies on membrane fouling have been made and reported, and it has been revealed, based on liquid chromatography organic carbon detection (LC-OCD), that biopolymer is the main foulant in the drinking water treatment process, in which the raw water is taken from a river or a dam. However, measurement by LC-OCD is time-consuming and costly. Therefore, continuous measurement of biopolymer concentration by LC-OCD is not feasible. For this reason, we have not been able to monitor biopolymer continuously and control membrane fouling. The purpose of this study is to find a new fouling index (FR) to control membrane fouling without measuring the biopolymer concentration. Then, we tried to find a correlation between biopolymer and other water components by a multiple regression analysis. As the result, we have suggested the new fouling index (FR) which consists of the sum of the fluorescence intensity within the Region III domain measured by excitation emission matrix (EEM) fluorescence spectroscopy and the concentration of dissolved organic carbon measured by the total organic carbon (TOC) measurement. TOC and EEM are measured easily and continuously. Thus, we can control membrane fouling by monitoring the FR continuously.

INTRODUCTION

The drinking water treatment system using a membrane has many advantages, such as complete removal of pathogenic bacteria and protozoa including Cryptosporidium, downsizing of the process equipment, and easy maintenance. Particularly, the use of the membrane system has been high in small-scale water purification plants. On the other hand, membrane fouling is a serious problem associated with this system, which causes an increase in operational power and frequency of chemical cleaning, and subsequently increases operating cost. Therefore, reduction in operating cost using chemical cleaning optimization has been considered (Yoo et al. 2018).

To control membrane fouling, we must identify the foulants, i.e. the materials that cause membrane fouling, and monitor them continuously. If we could measure the concentration of foulant in raw water continuously, we would be able to control membrane fouling by stopping the intake of raw water that has high fouling tendency or supplying chemicals to flocculate the foulants (Bogati et al. 2015).

In the case of a drinking water treatment system using a membrane, which takes raw water from surface sources such as rivers or dams, many researchers have reported that the main foulant must be a biopolymer related to biological metabolism on the membrane surface (Hallé et al. 2009; Peldszus et al. 2011; Tian et al. 2013). We have also succeeded in decreasing membrane fouling by reducing biopolymer concentration in raw water with a biological contact filter (Hasegawa et al. 2017).

Recently, biopolymer concentration has been measured accurately using liquid chromatography organic carbon detection (LC-OCD), in which the biopolymer is separated by a size-exclusion column (Chon et al. 2017). However, LC-OCD analysis is not suitable for continuous measurement of foulants in raw water, because it takes as long as 130 min per sample measurement. On the other hand, a total organic carbon (TOC) analyzer and excitation emission matrix (EEM) fluorescence spectroscopy can measure components in raw water in a short time at low cost; thus, it is more suitable for continuous measurement. However, it is reported that TOC does not have a clear correlation with membrane fouling (Kimura et al. 2014). Furthermore, though the correlation between membrane fouling and the components determined from EEM analysis using parallel factor analysis has been examined, a clear correlation between them has not been obtained (Shao et al. 2014; Yu et al. 2014).

In this study, we intend to establish a new fouling index that can be monitored continuously and predict membrane fouling. At first, we investigated the relationship between membrane fouling and raw water components for 13 water samples collected from nine rivers, and confirmed that biopolymer concentration had the strongest correlation with membrane fouling. Then, we tried to find the correlation between biopolymer concentration and the TOC value and EEM analysis using multiple regression analysis.

MATERIALS AND EXPERIMENTAL METHODS

Water sample

In this study, 40 water samples were collected from nine rivers (Akashi, Ikawa, Ikuta, Ishiya, Ina, Sumiyoshi, Toga, Mo, and Yodo) in the Hyogo and Osaka prefectures, Japan. They were collected from smooth flowing parts of the rivers using a bucket, filled in a 20 L plastic container, transported within several hours, and stored in a refrigerator without any pretreatment. Within 2 days, components of the samples were analyzed, and 13 water samples were selected by random sampling and fed to the filtration test equipment.

Filtration test of sampled water

In order to evaluate the fouling potential of the water samples, i.e. the tendency to cause membrane fouling, a filtration test was performed using mini microfiltration (MF) module test equipment (Figure 1). The sampled water was circulated through the mini MF column in which the mini MF module was set. The water temperature was controlled at 20 ± 0.5 °C during the test period, using a heater and a cooler. The permeated water was drawn through the mini MF module at a constant velocity of 1.5 m3/(m2·d). The mini MF module was backwashed with ultrapure water for 1 min every 30 min to reduce membrane fouling. The transmembrane pressure (TMP) was monitored until it exceeded 50 kPa. Specifications of the mini MF module and operating conditions are detailed in Figure 1.

Figure 1

Schematic drawing, specification, and operating conditions of the test equipment.

Figure 1

Schematic drawing, specification, and operating conditions of the test equipment.

The fouling potential of the sampled water was evaluated by the rate of increase of its membrane filtration resistance Rf (1/m·d), hereafter referred to as fouling rate (FR), which is defined by Equation (1): 
formula
(1)
 
formula
(2)
where ΔRt(tf) is the increment in membrane filtration resistance (1/m) during time tf, given by Equation (2); tf is the operating time of the filtration test (d); Δ(ΔP(tf)) is the increment in transmembrane pressure (Pa) during tf; J is the filtration flux (m/s), which is fixed at 1.5 m/s for this study; and μ is the viscosity (Pa·s) of the sampled water. Note that larger values of FR indicate a higher fouling potential of the sampled water.

Analytical method

Total organic carbon (TOC)

Prior to TOC measurement, sampled water was filtered through a 0.45-μm hydrophilic polytetrafluoroethylene membrane (DISMIC 13HP045AN, Advantec, Tokyo, Japan). Generally, TOC comprises dissolved organic carbon (DOC) and particulate organic carbon (POC). However, in this paper, POC was removed by filtration and only DOC was measured. Thus, the measured TOC is referred to as TOC-DOC.

The TOC of the sampled water was measured with a TOC analyzer (TOC-VCSH, Shimadzu Corporation, Kyoto, Japan). First, inorganic carbon was removed from the sampled water injected into the TOC analyzer as carbon dioxide at pH less than 3. Then, the filtered sampled water was burnt at 680 °C on a platinum catalyst. The burnt gas was cooled and dehumidified, and then the CO2 in the gas was detected with a non-dispersive infrared gas detector. Phthalate hydrogen potassium was used as a standard for calibration.

Excitation emission matrix (EEM) fluorescence spectroscopy

To characterize the dissolved organic matter (DOM) in the sampled water, EEM measurements were performed. Prior to the EEM measurements, all samples were filtered through the same filter used in the TOC measurement. EEM fluorescence was measured with an EEM spectrometer (Aqualog, Horiba Advanced Techno Co. Ltd, Kyoto, Japan). The excitation light was set from 220 to 800 nm and was irradiated for 1 s at intervals of 3 nm. When evaluating the three-dimensional fluorescence spectra, the spectra were divided into five domains and the sum of the fluorescence intensity within each domain was calculated (Chen et al. 2003). Hereafter, Reg. x indicates the sum of the fluorescence intensity within the Region x domain.

Liquid chromatograph organic carbon detector (LC-OCD)

LC-OCD measurements were performed to analyze the concentration of natural organic matter (NOM) in sampled water. Prior to the LC-OCD measurements, all samples were filtered through the same filter used in the TOC measurement. The organic carbon content was measured with a Model 8 DOC-Labor instrument (DOC-Labor, Karlsruhe, Germany). A hydrophilic weak cation exchange resin (TOYOPEARL HW-50, Tosoh, Tokyo, Japan) was used in the chromatographic column. The NOMs separated through the chromatographic column were oxidized by UV in a thin-film reactor. Then, the NOMs were converted to CO2 with an acidification liquid and removed from the feed. The concentration of NOMs was determined as the concentration of CO2 using a non-diffusion infrared detector. The mobile phase was a phosphorus acid buffer of pH 6.58, comprising sodium dihydrogen phosphate and disodium hydrogenphosphate (Sigma-Aldrich Japan, Tokyo, Japan). The acidification liquid was a phosphorus acid buffer of pH 1.5, comprising phosphoric acid and potassium peroxodisulfate (Sigma-Aldrich Japan, Tokyo, Japan). The mobile phase was injected at a flow rate of 1.1 mL/min, and the quantity of the sample injection was set at 1,000 µL. The total retention time was set at 130 min. The concentration of NOMs was calculated using the software chromCALC (DOC-Labor, Karlsruhe, Germany) customized to our system.

Multiple regression analysis

For multiple regression analysis, we used the regression analysis Excel pack from Microsoft Corporation (Washington, DC, USA). In the results of the multiple regression analysis, the effectiveness of the correlation was evaluated by a P-value (probability value) of 5%, that is, when the P-value was less than 5%, the correlation was estimated to have significance.

In the Excel pack, the coefficient of determination R2 is defined as Equation (3): 
formula
(3)
where yi indicates one of criterion value y, indicates an average value of y, and ŷi indicates a predictive value of yi. According to this definition, R2 can have a negative value as seen in Tables 1 and 2.
Table 1

The coefficient of determination of each component for fouling rate

TOC analysis
LC-OCD analysis
EEM analysis
ComponentsR2ComponentsR2ComponentsR2
TOC-DOC 0.475 LC-OCD-DOC 0.404 Reg. 1 0.268 
  BP 0.928 Reg. 2 0.017 
  HS 0.285 Reg. 3 0.083 
  BB 0.089 Reg. 4 0.015 
  LN 0.233 Reg. 5 0.023 
  LA −1.014   
TOC analysis
LC-OCD analysis
EEM analysis
ComponentsR2ComponentsR2ComponentsR2
TOC-DOC 0.475 LC-OCD-DOC 0.404 Reg. 1 0.268 
  BP 0.928 Reg. 2 0.017 
  HS 0.285 Reg. 3 0.083 
  BB 0.089 Reg. 4 0.015 
  LN 0.233 Reg. 5 0.023 
  LA −1.014   

R2: coefficient of determination; LC-OCD-DOC: sum of BP, HS, BB, LN and LA; BP: biopolymer, HS: humic substances, BB: building blocks, LN: low molecular weight nutrients, LA: low molecular weight acids.

Table 2

Relationship between biopolymer concentration and other components

TOC analysis
LC-OCD analysis
EEM analysis
ComponentsR2ComponentsR2ComponentsR2
TOC-DOC 0.387 LC-OCD-DOC 0.329 Reg. 1 0.441 
  BP 1.000 Reg. 2 0.105 
  HS 0.026 Reg. 3 0.103 
  BB 0.069 Reg. 4 0.153 
  LN −0.767 Reg. 5 0.152 
  LA 0.288   
TOC analysis
LC-OCD analysis
EEM analysis
ComponentsR2ComponentsR2ComponentsR2
TOC-DOC 0.387 LC-OCD-DOC 0.329 Reg. 1 0.441 
  BP 1.000 Reg. 2 0.105 
  HS 0.026 Reg. 3 0.103 
  BB 0.069 Reg. 4 0.153 
  LN −0.767 Reg. 5 0.152 
  LA 0.288   

R2: coefficient of determination.

RESULTS AND DISCUSSION

Components of river water which affect membrane fouling

The results of TOC, LC-OCD, and EEM measurements of water samples collected from the nine rivers are shown in the supplementary data (Table S1, available with the online version of this paper). The coefficient of determination (R2) values (Table 1), obtained from the linear regression analysis of each component for fouling rate, show that only biopolymer concentration correlates strongly with fouling rate. The second strongest correlation is with TOC-DOC; however, R2 of TOC-DOC (0.475) is much less than that for biopolymer concentration (0.928). Thus, if biopolymer concentration could be measured continuously, we could predict membrane fouling from the continuous monitoring of biopolymer concentration in raw water, and membrane fouling could be controlled automatically by chemical dosing or increasing the frequency of chemical cleaning before membrane fouling. However, unfortunately, the current LC-OCD analyzers cannot measure continuously, because it takes a long time (130 min) for one measurement and a dilution procedure is necessary to make the concentration lower than the upper limit of the LC-OCD measurement, and even then, the analysis cost is high. Thus, a new parameter is required, which correlates with membrane fouling and can be measured continuously.

New parameter instead of biopolymer concentration

The correlations between biopolymer concentration and other components of sampled water analyzed by linear regression (Table 2) show that no component correlates with biopolymer concentration, because the R2 value of each parameter is lower than 0.5. Thus, we need to consider the correlation between biopolymer concentration and a combination of other components.

The concentration of biopolymer, [BP], is given by Equation (4): 
formula
(4)
where HS indicates humic substances, BB building blocks, LN low molecular weight nutrients, and LA low molecular weight acids; [ ] indicates the concentration.
We conclude that [LC-OCD-DOC] has a correlation with [TOC-DOC], because both values show the DOC concentration of the same sampled water, even if the analytical method is different. The result of linear regression analysis between [LC-OCD-DOC] and [TOC-DOC] (Figure 2) shows that [TOC-DOC] has a strong correlation with [LC-OCD-DOC]. This means that the DOM of river water can be detected with an LC-OCD analyzer using a chromatographic column with a hydrophilic weak cation exchange resin. As a result, Equation (4) can be rewritten as Equation (5): 
formula
(5)
Figure 2

Correlation between [TOC-DOC] and [LC-OCD-DOC].

Figure 2

Correlation between [TOC-DOC] and [LC-OCD-DOC].

In Equation (5), it is reasonable to consider that [HS], [BB], [LN], and [LA] have a correlation with EEM results, because EEM also measures DOM by dividing it into five domains. Thus, we attempt multiple regression analysis for FR by rewriting Equation (5) as Equation (6): 
formula
(6)
where a, b, c, d, e, f, and g are constants.
The results of the multiple regression analysis for FR values, shown in Table S1, with Equation (6) (Table 3(a)) show that Equation (6) fits with FR very well (R2 = 0.975). However, the P-values of Regs. 2, 4, and 5 are higher than 0.05. Thus, we retried the analysis using Equation (7): 
formula
(7)
where h, i, j, and k are constants.
Table 3

Multiple regression analysis for FR

TermCoefficientP-valueR2
(a) With Equation (6) 
 TOC-DOC 2.77 6.16 × 10−5 0.975 
 Reg. 1 9.99 × 10−6 0.00742  
 Reg. 2 −6.70 × 10−7 0.821  
 Reg. 3 −4.26 × 10−6 0.0331  
 Reg. 4 −4.28 × 10−6 0.108  
 Reg. 5 5.26 × 10−7 0.326  
 Intercept −1.48 0.000345  
(b) With Equation (7) 
 TOC-DOC 3.01 4.27 × 10−5 0.910 
 Reg. 1 1.59 × 10−6 0.412  
 Reg. 3 −4.31 × 10−6 0.000110  
 Intercept −1.34 0.00132  
(c) With Equation (8) 
 TOC-DOC 3.10 1.31 × 10−5 0.903 
 Reg. 3 −4.17 × 10−6 6.17 × 10−5  
 Intercept −1.34 0.000924  
TermCoefficientP-valueR2
(a) With Equation (6) 
 TOC-DOC 2.77 6.16 × 10−5 0.975 
 Reg. 1 9.99 × 10−6 0.00742  
 Reg. 2 −6.70 × 10−7 0.821  
 Reg. 3 −4.26 × 10−6 0.0331  
 Reg. 4 −4.28 × 10−6 0.108  
 Reg. 5 5.26 × 10−7 0.326  
 Intercept −1.48 0.000345  
(b) With Equation (7) 
 TOC-DOC 3.01 4.27 × 10−5 0.910 
 Reg. 1 1.59 × 10−6 0.412  
 Reg. 3 −4.31 × 10−6 0.000110  
 Intercept −1.34 0.00132  
(c) With Equation (8) 
 TOC-DOC 3.10 1.31 × 10−5 0.903 
 Reg. 3 −4.17 × 10−6 6.17 × 10−5  
 Intercept −1.34 0.000924  
Table 3(b) shows the results of the multiple regression analysis for FR with Equation (7). Equation (7) fits very well with FR, though the R2 for Equation (7) (0.910) is slightly less than that for Equation (6) (0.975). However, the P-value of Reg. 1 is higher than 0.05. Therefore, we retried the analysis using Equation (8): 
formula
(8)
where l, m, and n are constants.
Table 3(c) shows the results of the multiple regression analysis for FR with Equation (8). It is found from Table 3(c) that the P-value of each term is less than 0.05. In addition, R2 is 0.903, which means Equation (8) fits with FR well enough, even though R2 at 0.903 is a little lower than 0.975 for Equation (6) and 0.910 for Equation (7). Equation (8) is rewritten as Equation (9) by inserting the coefficients shown in Table 3(c) into Equation (8): 
formula
(9)

Figure 3 shows the correlation between the experimental FR and the estimated FR with Equation (9). It is clear that FR is estimated well with Equation (9).

Figure 3

Correlation between FR obtained experimentally and the estimated FR in Equation (9).

Figure 3

Correlation between FR obtained experimentally and the estimated FR in Equation (9).

It will be possible to predict membrane fouling by consecutive measurement of the TOC and EEM and estimation of the FR with Equation (9). Figure 4 shows a flowchart of a fouling control system that uses an arithmetic unit for fouling prediction.

Figure 4

Image of fouling control system using arithmetic unit for fouling prediction.

Figure 4

Image of fouling control system using arithmetic unit for fouling prediction.

Table 4

R2 values in correlation between each component of EEM and LC-OCD analysis

LC-OCD analysisEEM analysis
Reg. 1Reg. 2Reg. 3Reg. 4Reg. 5
BP 〇 0.441  0.105  0.103  0.153  0.152 
HS  0.601  0.798 〇 0.934  0.777  0.877 
BB  0.718  0.883 〇 0.936  0.882  0.908 
LN 〇 0.630  0.328  0.337  0.395  0.373 
LA 〇 0.137  0.106  0.100  0.126  0.094 
LC-OCD analysisEEM analysis
Reg. 1Reg. 2Reg. 3Reg. 4Reg. 5
BP 〇 0.441  0.105  0.103  0.153  0.152 
HS  0.601  0.798 〇 0.934  0.777  0.877 
BB  0.718  0.883 〇 0.936  0.882  0.908 
LN 〇 0.630  0.328  0.337  0.395  0.373 
LA 〇 0.137  0.106  0.100  0.126  0.094 

〇: Component of EEM analysis that has the strongest correlation with each component of LC-OCD analysis.

R2: coefficient of determination.

Discussion of Equation (9)

From Equation (9), it is found that only Region 3 of the EEM has a correlation with FR. Other regions of the EEM have no correlation with FR. Table 4 shows the correlation of linear regression between each component of EEM analysis and LC-OCD analysis. Humic substances ([HS]) and building blocks ([BB]) have a very strong correlation with Reg. 3, and low molecular weight nutrients ([LN]) and acids ([LA]) with Reg. 1. Therefore, Equation (5) can be rewritten as Equation (10), which is similar to Equation (7):

 
formula
(10)
where p, q, r, s and t are constants.

As seen in Table S1, [HS] + [BB] is higher than [LN] + [LA]. Thus, the contribution of Reg. 3 to the difference between [LC-OCD-DOC] and [BP] is probably stronger than that of Reg. 1. Therefore, FR correlates with only [TOC-DOC] and Reg. 3, as given by Equation (9).

Of course, soluble BOD and COD may be usable in substitution for TOC-DOC because they are associated with TOC-DOC. But in the case of applying enzymatic reaction or ultraviolet absorbance methods for the consecutive measurement of BOD or COD, they would be unsuitable because the value is influenced by water composition. Also regarding other water quality indexes, for example, the Effluent Quality Index (EQI) explained by Pathnayake & Tanyimboh (2015), as it includes various compositions other than dissolved organic matter, it will be unusable in substitution for TOC-DOC.

CONCLUSION

This study examined a foulant monitoring method with low cost and easy operation. We found good correlation between fouling rate and two components of the sampled water, viz. TOC and Reg. 3, measured by EEM. The following equation was derived to predict fouling rate: 
formula

TOC-DOC and Reg. 3 can be measured continuously and with little effort. By measuring TOC-DOC and Reg. 3, and estimating FR using the above equation, we can predict the tendency of membrane fouling and control it.

ACKNOWLEDGEMENT

This work was financially supported by the Kansai bureau of economy, trade, and industry (METI-Kansai), Japan.

AUTHOR CONTRIBUTIONS

Susumu Hasegawa and Taro Miyoshi conceived and designed the experiments; Susumu Hasegawa and Taro Miyoshi performed the experiments; Susumu Hasegawa, Taro Miyoshi, and Ryosuke Takagi analyzed the data; Susumu Hasegawa, Ryosuke Takagi, and Hideto Matsuyama contributed to writing the manuscript.

REFERENCES

REFERENCES
Bogati
R.
,
Goodwin
C.
,
Marshall
K.
,
Leung
K. T.
&
Liao
B. Q.
2015
Optimization of chemical cleaning for improvement of membrane performance and fouling control in drinking water treatment
.
Sep. Sci. Technol.
50
,
1835
1845
.
Chen
W.
,
Westerhoff
P.
,
Leenheer
J. A.
&
Booksh
K.
2003
Fluorescence excitation–emission matrix regional integration to quantify spectra for dissolved organic matter
.
Environ. Sci. Technol.
37
,
5701
5710
.
Hallé
C.
,
Huck
P. M.
,
Peldszus
S.
,
Haberkamp
J.
&
Jekel
M.
2009
Assessing the performance of biological filtration as pretreatment to low pressure membranes for drinking water
.
Environ. Sci. Technol.
43
,
3878
3884
.
Hasegawa
S.
,
Iwamoto
T.
,
Miyoshi
T.
,
Onoda
S.
,
Morita
K.
,
Takagi
R.
&
Matsuyama
H.
2017
Effect of biological contact filters (BCFs) on membrane fouling in drinking water treatment systems
.
Water
9
,
981
.
Pathnayake
U. S.
&
Tanyimboh
T. T.
2015
Evolutionary multi-objective optimal control of combined sewer overflows
.
Water Resour. Manage.
29
,
2715
2731
.

Supplementary data