Combining LC-OCD Analysis with Design-of-Experiments Methods to

Optimize an Advanced Oxidation Process for the Treatment of

Industrial Wastewater

Kimia Aghasadeghi,1 Matthew Csordas,1 Sigrid Peldszus2 and David R. Latulippe1*

1. Department of Chemical Engineering, McMaster University, 1280 Main Street West, Hamilton, ON, L8S 4L7, Canada
2. Department of Civil and Environmental Engineering, University of Waterloo, 200 University Ave. West, Waterloo, ON, N2L 3G1, Canada

Advanced oxidation (AO) is widely used as a pre-treatment and/or polishing step for the treatment of wastewater from industrial processes and
the destruction of particular contaminants in water sources. It has a high treatment efficacy for many different compounds and thus is ideally
suited as a treatment technology for specialized facilities that receive shipments of wastewater from networks of industrial, manufacturing, and
commercial facilities. The primary challenge is how to optimize the process because bulk measurements of organic content (e.g. TOC) give no
information about the specific composition and specialized advanced analytical techniques (e.g. LC-MS) are unsuitable due to the complex
composition. In this study, a novel combination of design-of-experiments (DOE) methods and LC-OCD analysis was used with actual wastewater
samples in order to optimize the AO treatment conditions in terms of chemical reagent concentrations, develop statistical models of the process,
and identify potential mechanisms of COD removal. A significant variation in organic content removal was obtained over the range of conditions
tested in the DOE method. For example, the percent removal of organic contaminants in the one wastewater sample varied from a low of 36 %
to a high of 82 %. Most importantly, it was found that the treatment performance differed quite significantly for wastewater samples of different
composition. The results presented in our study prove the need to dynamically optimize the AO treatment conditions for wastewater sources
of different origins. Furthermore, by the application of the LC-OCD analysis a step-by-step mechanism of COD removal was postulated.

Keywords: wastewater treatment, advanced oxidation, LC-OCD, chemical oxygen demand, Fenton process

INTRODUCTION those that focus on the destruction of all organic components in

the wastewater from a single specific source such as pharmaceu-

A
dvanced oxidation (AO) is a non-selective treatment tical manufacturing,[8,9] pulp and paper processing,[10] food
process that uses the strong oxidizing power of manufacturing,[11] or textile manufacturing.[12] These studies
chemical radicals for complete or partial mineralization typically use bulk analytical measurements of the organic
of organic contaminants.[1] AO processes are frequently used contaminants (including chemical oxygen demand (COD),[8–11]
in industrial wastewater treatment for breaking down the non- biochemical oxygen demand (BOD),[9] or total organic carbon
biodegradable contaminants as part of the pre-treatment and (TOC)[9,12]) and because the wastewater composition is fairly
‘polishing’ steps of biological treatment processes.[1] There are consistent from the single source, it is rather straightforward
multiple types of AO processes including those based on to optimize the AO process conditions in terms of treatment
Fenton’s reaction such as photo-Fenton reaction,[2,3] electro- efficacy. The second category comprises studies that focus
Fenton reaction,[4] and sono-Fenton reaction.[5] The Fenton on the destruction of a particular component (e.g. pesticide/
process is based on the generation of hydroxyl radicals from herbicide,[13,14] synthetic dye,[15] chlorophenol[16]) that is present
hydrogen peroxide at acidic pH in the presence of ferrous ions.[2] in a particular wastewater source. These studies monitor the
The principal reaction of the Fenton process is shown below as destruction of the known component as a function of the AO
Equation (1): process conditions using advanced analytical techniques such as
gas chromatography,[13] gas chromatography-mass spectrometry
Fe2þ þH2 O2 ! Fe3þ þ OH þ OH ð1Þ (GC-MS),[16] high-performance liquid chromatography
(HPLC),[14,15] or liquid chromatography-mass spectrometry (LC-
MS).[14,15]
However, there is actually a total of nine reactions involved in Another common practice in the field of environmental
the process.[5] The simplicity of the Fenton process reaction, in processes is the operation of specialized wastewater treatment
addition to no energy input requirements and short reaction
time,[6] has made it an attractive ‘polishing’ treatment step. Also, it
is a robust process that can achieve a high treatment efficacy
for many different compounds.[7] The main disadvantage of the * Author to whom correspondence may be addressed.
Fenton process is the high operational costs due to the use of E-mail address: [email protected]
expensive reagents.[5,6] Thus, there exists a strong need to Can. J. Chem. Eng. 9999:1–10, 2017
optimize the reaction conditions to obtain the best treatment for © 2017 Canadian Society for Chemical Engineering
DOI 10.1002/cjce.22880
a given cost. Supplementary material is available in the online journal.
Previous optimization studies of AO-based processes can be Published online in Wiley Online Library
grouped into two general categories. The first category includes (wileyonlinelibrary.com).

VOLUME 9999, 2017 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 1

plants that accept and treat tanker truck shipments of EXPERIMENTAL
wastewater from multiple industrial, manufacturing, and
commercial facilities—those facilities do not do any on-site Wastewater Sample Source
treatment of the wastewater generated and thus they cannot
Two industrial wastewater samples (hereafter referred to as
discharge the untreated wastewater. Maintaining the treatment
wastewater 1 and wastewater 2) were provided by the research
process efficacy to meet discharge regulations is very challeng-
team at Aevitas (Brantford, ON, Canada). Both samples had
ing at these specialized treatment facilities as the composition
undergone a preliminary treatment for partial removal of large
of the incoming feed to the process changes with each
particulate matter and insoluble oils. The Aevitas plant receives
shipment. AO is a promising technology for application in
and treats loads of wastewater that are generated by a network of
these facilities as it has a high treatment efficacy for many
many hundreds of industrial and commercial facilities. In order to
different compounds[7] but the challenge is to optimize the
ensure that the two samples were of significantly different
process for different shipments. Surprisingly this field of
composition, they were collected from the plant approximately
wastewater treatment has attracted little attention. To the
one month apart. Exact details on the original source of the two
best of our knowledge, only one previous study has optimized
samples are not available. Basic properties of the two samples are
the AO conditions for wastewater loads from multiple sources.
listed in Table 1; the initial COD (CODi) was measured using
Bianco et al.[7] studied COD removal via the Fenton reaction
HACH high range COD vials as per the manufacturer’s instruc-
on wastewater samples which had a wide range of initial COD
tions and the pH was measured using a sympHonyTM SB20 meter
and were very diverse in composition. Those samples were
(VWR). The samples were stored at 4 8C and the experiments
obtained from an industrial wastewater treatment plant that
were conducted less than one week after sample collection.
treated wastewater from different industries. The percent
removal of COD was determined for different ratios of the Experimental Procedure
reaction reagents.
A schematic summary of the experimental procedure is shown in
One of the challenges faced in the optimization of AO
Figure 1. The AO experiments were conducted at room tempera-
processes for application in wastewater treatment plants with
ture in 150 mL glass beakers using 50 mL of the wastewater
dynamically varying incoming loads is the initial characteriza-
sample for each experimental condition. Each treatment condition
tion of the wastewater composition. Neither of the approaches
in the DOE (described in detail below) was done in duplicate; the
in the two categories of studies mentioned above are appropri-
vessel from which the sample was withdrawn was continuously
ate; bulk measurements (e.g. COD, BOD, TOC) tell the
stirred to ensure that a reproducible sample was aliquoted into
researcher nothing about the initial composition or changes
each beaker. The first step was to lower the pH of the as-received
in composition post-treatment and it is not feasible to use
wastewater sample to 3.0 ( 0.1) via the addition of 1 mol/L
advanced analytical techniques (e.g. GC-MS, HPLC, LC-MS)
hydrochloric acid (Sigma Aldrich); it has been previously reported
to identify specific contaminants due to the high level of
that the optimal results are achieved at this low pH condition.[5]
contamination and complex matrices of the incoming waste-
The COD removal efficacy of the Fenton’s reaction was
water. Thus, an alternative method was sought for this study.
determined using the two independent factors S and R that are
Liquid chromatography-organic carbon detection (LC-OCD) is
defined by Equations (2) and (3) below:
an analytical tool that was developed approximately 25 years
ago to fill an ‘analytical gap’ in characterizing the amount of
H2 O2 Molar Concentration ðmol=m3 Þ
natural organic matter (NOM) in water/wastewater sources. S: ð2Þ
The method uses a size exclusion column combined with an Wastewater COD ðmg=LÞ
organic carbon detector, organic nitrogen detector, and a UV
absorption detector to fractionate samples according to their H2 O2 Molar Concentration ðmol=m3 Þ
R: ð3Þ
size and hydrophobicity.[17] It has been applied to study NOM Fe2þ Molar Concentration ðmol=m3 Þ
fractions in surface water,[18] municipal wastewater,[19] and sea
water,[20] and the efficacy of different NOM removal treat- As shown in Figure 1, a central composite design (CCD) in the
ments.[19,20] Additionally, it has been used to study the form of a 22 full factorial with two extra face points at (1,0) was
treatment of recalcitrant organic matter in paper mill efflu- used in this study; the axial points were located at (0, a) and
ent,[21] and also the effect of photo-oxidation on NOM in cooling (a, 0) where a, the coded unit distance of the axial points from
water.[22] These studies have demonstrated the valuable insight the centre point, was chosen to be 1.4. Thus, a total of 11 different
of LC-OCD analysis towards an improved understanding of the treatment conditions were investigated for both wastewater
complex relationship between sample composition and treat- samples that were studied in this work. The coded and real
ment efficacy that is not possible via conventional bulk values of the S and R factors and their associated molar
measurements such as COD and TOC. concentrations of reagents used in different treatment conditions
Given the usefulness of the LC-OCD method as a characteriza-
tion tool, it was used in this work to study the effect of
sample composition on Fenton-based COD removal efficacy
for two industrial wastewater samples that are sufficiently
distinct in composition. A statistical design-of-experiment Table 1. Properties of the as-received industrial wastewater samples
(DOE) strategy was used to vary the Fenton reaction conditions. from Aevitas
It was found that the combination of LC-OCD and DOE methods Sample COD (mg/L) pH
is an effective strategy for optimizing AO treatment efficacy
by providing information on the different mechanisms of Wastewater 1 2700 8.4
COD removal that take place during the treatment. Wastewater 2 4000 6.0

2 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 9999, 2017

Figure 1. Schematic of the experimental workflow for the wastewater samples including the various analytical tests that were performed.

are presented in Table 2. The coded values represent the real out into a sludge, and it deactivated any residual H2O2 that could
values after being centred and scaled. potentially interfere with the subsequent analysis steps. The
A fresh solution of 50 g/L ferrous sulphate heptahydrate samples were again left for 1 h to allow the iron precipitates to
(Sigma-Aldrich) in ultrapure water (obtained from a Millipore settle to the bottom of the beaker. Multiple aliquots of the clear
Milli-Q system) was prepared for each set of oxidation tests. The supernatant layer were used for the LC-OCD analysis (Figure 1)
ferrous sulphate heptahydrate solution was added to each beaker and the final COD measurement (CODf) using the same HACH
based on the corresponding S and R values—refer to Table 2. high range COD vials; the COD measurements were done in
After the addition of ferrous sulphate heptahydrate solution, the duplicate. The dilution effect associated with the addition of
hydrogen peroxide solution (35 g/g from BDH) was added to each different volumes of the ferrous sulphate heptahydrate and
beaker to begin the reaction. The volume of hydrogen peroxide hydrogen peroxide solutions was considered in the CODf
added to each beaker was based on the corresponding S value; determination. The percentage removal of COD, as defined by
refer to Table 2 for the molar concentrations of H2O2 and Fe2þ Equation (4), was chosen as the objective function in the DOE
based on the moles of reagent available per volume of the sample. method.
The reaction time was kept fixed at 1 h followed by the addition of
50 g/g sodium hydroxide solution (Sigma Aldrich) to increase CODi  CODf
% COD Removal ¼  100 ð4Þ
the pH to greater than 8. This step was required for three reasons: it CODi
stopped the oxidation reaction, it caused the iron to precipitate

VOLUME 9999, 2017 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 3

 Table 2. Coding values and associated molar concentrations of reagents used in different treatment conditions

S (H2O2/COD) R (H2O2/Fe2þ) Wastewater 1 Wastewater 2

Coded Real Value Coded Real Value H2O2 Fe2þ H2O2 Fe2þ
Value (mol/m3)/(mg/L) Value (mol/m3)/(mol/m3) (mol/m3) (mol/m3) (mol/m3) (mol/m3)

-1 0.1 0 10 270 27 400 40

þ1 0.2 þ1 15 540 36 800 53
-1 0.1 þ1 15 270 18 400 27
þ1 0.2 0 10 540 54 800 80
-1 0.1 -1 5 270 54 400 80
þ1 0.2 -1 5 540 108 800 160
0 0.15 0 10 405 41 600 60
0 0.15 þ1.4 17 405 24 600 35
0 0.15 -1.4 3 405 135 600 200
-1.4 0.08 0 10 216 22 320 32
þ1.4 0.22 0 10 594 59 880 88

Liquid Chromatography  Organic Carbon Detection (LC-OCD) different, especially at higher retention time (RT) values. For
The supernatant of the treated samples was collected in a plastic example, the large peak at a retention time of approximately
syringe and then passed through a 0.45 mm Supor1 polyethersul- 80 min for the wastewater 2 sample constitutes about 60 % of its
phone syringe filter (Pall Corporation) into a TOC clean glass vial. overall DOC content while no such peak was measured for the
The filters were pre-rinsed by passing approximately 20 mL of wastewater 1 sample. Thus, we felt that the two samples were
ultrapure water through them and then dried by passing air good candidates for the optimization study of an AO treatment
through them. Due to their high dissolved organic carbon (DOC) process. The original development of the LC-OCD analytical
content all samples had to be diluted 50–500 times with ultrapure method focused on water samples from natural sources and thus
water before the LC-OCD analysis. Samples were analyzed using the regular occurrence of specific peak patterns in the resulting
the LC-OCD system (from DOC-LABOR Huber) at the University of chromatograms were assigned to particular organic matter
Waterloo. A detailed description of the working principles and fractions (e.g. biopolymers, humic substances, low molecular
operation of the LC-OCD instrument has been given in previous weight acids).[17] However, those categories are not expected to be
works.[17,23] Briefly, the LC-OCD instrument uses a size exclusion suitable for the industrial wastewater samples used in this study
column combined with an organic carbon detector (OCD), organic given their radically different sources. Therefore, we relied
nitrogen detector (OND), and a UV absorption detector (UVD) to partially on the LC-OCD calibration with PEG standards. The
fractionate samples according to their size and hydrophobicity. retention times of three of the PEG standards are shown on the
Organic carbon detection is achieved via the Gra €ntzel reactor, a x-axis of Figure 2 as a reference while the complete calibration
UV thin-film reactor in which organic carbon is converted to data (for molecular weight range from 0.2 to 20 kg/mol) is
carbon dioxide by UV irradiation. The carbon dioxide produced is
measured by high-sensitivity infrared spectrometry. The UV light
exposure and residence time in the reactor are sufficient for
quantitative organic carbon detection.[17] In this study only the
chromatograms from the OCD were analyzed. Polyethylene glycol
(PEG) standards from Sigma Aldrich were used to characterize the
fractionation range of the size exclusion column according to the
OCD signal; PEG is a linear chain polymer with the following
chemical structure: H  (O  CH2  CH2)n  OH. The chromato-
grams were analyzed using the ChromCALC software that is
provided by the manufacturer of the LC-OCD instrument. The
software accounts for the dilution factor of each sample. The
dilution effect associated with the addition of different volumes
of the ferrous sulphate heptahydrate and hydrogen peroxide
solutions was also considered in the presentation of the
chromatograms.

RESULTS AND DISCUSSION

Figure 2. Comparison of the LC-OCD chromatograms for the two
Comparison of LC-OCD Chromatograms for the Industrial wastewater samples that were used in this study; the solid and dashed lines
Wastewater Samples correspond to wastewater 1 and wastewater 2 respectively. Dilution factors
are accounted for in the plot. The peak retention times associated with the
The organic carbon detector (OCD) chromatograms from the molecular weight (kg/mol) of three PEG standards are shown on the x-axis
LC-OCD analysis of the two as-received industrial wastewater for comparison; note that a total of eleven PEG standards were actually
samples are shown in Figure 2. The chromatograms are quite used to calibrate the LC-OCD system.

4 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 9999, 2017

provided in the Supplementary Material. The retention time of
Table 3. Coefficients and 95 % confidence intervals for the full second
the dominating peak in wastewater 2 is outside the PEG order regression model of the COD removal results shown in Figure 3
calibration range and based on the calibration it can only be
concluded that the apparent molecular weight for this peak is Wastewater 1 Wastewater 2
smaller than 0.2 kg/mol. Ruhl and Jekel[24] reported that the
retention time of low molecular weight organics in LC-OCD Coefficient Value 95 % CI Value 95 % CI
analysis is not governed by the molecular weight but by the Intercept 63.12 (59.02, 67.21) 47.80 (44.74, 50.86)
chemical structure and functional groups. S 12.60 (11.06, 14.15 7.58 (6.43, 8.74)
R –12.54 (–14.27, 1.82) –0.81 (–2.1, 0.48)
Statistical Analysis of COD Removal via AO Treatment SR 2.12 (–0.31, 4.54) 1.75 (–0.07, 3.57)
The percent removal of COD was calculated for each of the 11 S2 –2.71 (–5.48, 0.05) –2.58 (–4.65, 0.51)
treatment conditions in the DOE method with the results shown in R2 –0.65 (–3.11, 1.82) 0.66 (–1.18, 2.51)
Figure 3. For wastewater 1 (Figure 3a), the different treatment
conditions resulted in a wide range of percent COD removals with
the lowest value (36 %) obtained at the (-1,þ1) coded condition
and the highest value (82 %) obtained at the (þ1,-1) coded on COD removal. The best treatment conditions in terms of COD
condition. For wastewater 2 (Figure 3b), there was less variation removal were obtained at higher S values (i.e. higher H2O2
in the percent COD removals achieved compared to wastewater 1. concentrations) and lower R values (i.e. lower H2O2/Fe2þ ratios).
For example, the values ranged from 31 % (at the (-1.4,0) coded This observation is in good agreement with previous studies.[26,27]
condition) to just 54 % (at both the (þ1,0) and (þ1.4,0) coded The second order S term and the interaction term (i.e. S2 and SR)
conditions). Additionally, significantly different percent COD have confidence intervals that span zero asymmetrically and
removals were achieved for the two wastewaters at the same thus were kept in the model as per recommended practice. The
treatment conditions. For example, at the (0,-1.4) coded condition second order R term (i.e. R2) was determined to be statistically
the percent COD removal was 79 % for wastewater 1 but only insignificant based on its relatively symmetric confidence interval.
48 % for wastewater 2. Comparison of the results at the same The finalized regression model is shown in Table 4 with the
treatment conditions for the two samples indicates that the associated contour plots based on those model coefficients
optimum treatment conditions are significantly different for the displayed in Figure 3a.
two samples. As shown in Table 3, for the wastewater 2 sample, the first
In order to develop a deeper understanding of the effect of the order term of S has a significant effect on COD removal and best
process variables, a second order regression model was applied to treatment conditions in terms of COD removal were obtained at
both datasets and the suitability of each term in the model was higher S values (i.e. higher H2O2 concentrations). However, the
determined using R software.[25] The statistical significance of the first order term of R has a confidence interval that spans zero
coefficients were determined based on their 95 % confidence asymmetrically and its magnitude is much smaller compared to
interval (CI). As shown in Table 3, for the wastewater 1 sample that of S which indicates that the effect of H2O2/Fe2þ ratio is
both of the first order terms (i.e. S and R) have a significant effect not as significant as the effect H2O2 concentration on the COD

a b H2O2 (mol/m3)/COD (mg/L)

H2O2 (mol/m3)/COD (mg/L)
0.075 0.1 0.125 0.15 0.175 0.2 0.225 0.075 0.1 0.125 0.15 0.175 0.2 0.225
1.5 20 % 30 % 40 % 50 % 55 % 60 % 17.5 1.5 30 %
40 % 50 % 55 %
17.5

42 % 51 %
36 % 62 % 65 % 34 % 51 %
1 15 1 15
H2O2 (mol/m3)/Fe2+ (mol/m3)

70 % H2O2 (mol/m3)/Fe2+ (mol/m3)

0.5 12.5 0.5 12.5
R (Coded Value)

R (Coded Value)

39 % 43 % 66 % 73 % 76 % 31 % 37 % 48 % 54 % 54 %
0 10 0 10
53 %

-0.5 80 % 7.5 -0.5 7.5

64 % 82 % 43 % 53 %
85 %
-1 5 -1 5

79 % 48 % 51 %

-1.5 2.5 -1.5 2.5

-1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5
S (Coded Value) S (Coded Value)

Figure 3. Contour plots of percent removal of COD as a function of the S and R coded values in the DOE design (22 full factorial with two extra face points
at  1,0) for wastewater 1 (panel a) and wastewater 2 (panel b). The associated contour plots were generated based on the model coefficients in Table 4
with the contour line values shown in italics. The bottom x-axis and left side y-axis display the coded S and R values respectively; the top x-axis and right side
y-axis display the ratios of reagents. The actual molar concentrations are given in Table 2.

VOLUME 9999, 2017 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 5

 obtained for the wastewater 2 sample indicate that the R value
Table 4. Coefficients and 95 % confidence intervals for the final
regression model of the COD removal results shown in Figure 3
had an effect at low S levels. This is likely due to differences in
the composition of the two wastewater samples as demonstrated
Wastewater 1 Wastewater 2 via the LC-OCD analysis (refer to results shown in the next
section).
Coefficient Value 95 % CI Value 95 % CI The statistical analyses presented above indicate that a universal
Intercept 60.27 (59.83, 64.69) 48.68 (46.85, 50.51) model cannot be applied to predict treatment efficacy for
S 12.61 (11.10, 14.11) 7.58 (6.45, 8.72) wastewaters from different sources. Bianco et al.[7] addressed
R –12.54 (–14.23, 10.86) 0.81 (2.08, 0.46) this issue by developing different models for different COD
SR 2.12 (0.25, 4.49) 1.75 (0.04, 3.54) ranges. However, the two wastewater samples used in this study
S2 2.29 (4.46, 0.11) 3.02 (4.66, 1.38) were in the same COD range and therefore LC-OCD analysis of the
treated samples was integrated into the experimental workflow.

Comparison of LC-OCD Chromatograms for Post-AO Treatment

removal results. Similar to the wastewater 1 sample, the second
order term of S and the interaction term (i.e. S2 and SR) have The key process steps involved in the Fenton reaction are to first
confidence intervals that span zero asymmetrically and the second adjust the pH to acidic conditions (pH  3 to 4) and then to add the
order term of R (i.e. R2) has a statistically insignificant effect. hydrogen peroxide and ferrous sulphate. The first step is important
Therefore, similar to wastewater 1 sample, a revised regression as it enables the activation of hydroxyl radicals and prevents the
model was applied with the results shown in Table 4; the precipitation of ferric hydroxide out of the solution.[28] A series of
associated contour plots based on those model coefficients are control experiments involving both of those process steps were
displayed in Figure 3b. The main difference between the conducted to determine their possible effects on the LC-OCD
regression models obtained for the two wastewater samples is analysis.
the magnitude of the first order term of R. The experimental results Both of the as-received wastewater samples were subjected to
the exact same pH-adjustment step (i.e. decrease pH to 3 for 1 h
then raise the pH to above 8) that was used for each treatment
condition in the DOE setup. As shown in Figure 4, the LC-OCD
2000
a: Wastewater 1
analysis showed that the two samples had very different
responses to the pH-adjustment step. For the wastewater 1
1 sample (shown in Figure 4a), the positions of the peaks were the
1500 same but the sizes of the peaks increased. Specifically, the size of
OCD Signal (au)

peak 1 increased by approximately 40 %; it is hypothesized that

this result is due to the transformation of particulate matter into
dissolved DOC.[22] For the wastewater 2 sample (shown in
1000 Figure 4b), the change in the chromatogram was much more
drastic. Specifically, the pH-adjustment step transformed the one
large peak at a retention time of 80 min into two peaks at lower
500 retention times of approximately 51 min (denoted as 1) and
62 min (denoted as 2). It is hypothesized that this result is
due to the aggregation of small molecules as has been
previously reported for humic substances.[29] Additionally, the
0
pH-adjustment step resulted in a 15 % increase in the DOC
30 40 50 60 70 80 90 100
content of the sample. Thus, it is quite apparent that the pH
5000
1 b: Wastewater 2 adjustment step affects different samples in different ways;
future work will explore in greater detail the exact mechanism of
pH-adjustment effects. All of the subsequent comparisons of the
4000 different AO treatment conditions in the DOE setup will be
made to the control pH-adjusted samples in order to isolate the
effects of the actual AO treatment condition.
OCD Signal (au)

3000 Control experiments were also carried out to determine whether

hydrogen peroxide and ferrous sulphate cause any signal in the
2000
2 LC-OCD chromatograms. This was achieved by obtaining the LC-
OCD chromatograms for each of these chemicals at the concen-
trations used in the centre condition (i.e. S ¼ 0.15, R ¼ 10) of
1000 wastewater 1 respectively. It was observed that the signal caused
by these chemicals in the OCD chromatograms is negligible. The
final control experiment was the combination of all the treatment
0
chemicals (i.e. hydrogen peroxide, ferrous sulphate, hydrogen
20 30 40 50 60 70 80 90 100
chloride, and sodium hydroxide) to ultrapure water at the same
-1000 Retention Time (min) concentrations used in the centre condition of wastewater 1
sample. A comparison of the OCD chromatogram of the control
Figure 4. Comparison of LC-OCD chromatograms for the as-received experiment with the chromatogram of the centre condition of
samples (dashed line) and post-pH adjustment of the as-received samples wastewater 1 sample is shown as Figure S-1 in the Supplementary
(solid lines) for both wastewater 1 (panel a) and wastewater 2 (panel b). Material. It was found that the AO chemicals alone produced a

6 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 9999, 2017

small peak (identified as 3) which also appears in the chromato- corresponds to a specific S and R condition as indicated in the
grams of the treated samples. panel legend; the percent COD removal is also given in the panel
As shown in Figure 1, the supernatants from all 11 treatment legend. The six panels are arranged strategically to match the
conditions in the DOE method were subjected to LC-OCD analysis. arrangement of the S ¼ 0.1 and S ¼ 0.2 conditions in the DOE
However, we chose to present the LC-OCD chromatograms for setup that was originally presented in Figure 3. Note that these six
a subset of the full experimental design. The chromatograms for conditions actually span the full range of percent COD removal
the wastewater 1 sample are shown in Figure 5. Each panel (36 % to 82 %) that was found over the entire DOE method.

2 500 S = 0.1 a 2 500 S = 0.2 b

R = 15 R = 15
2 000 2 000
% COD 2 % COD
2
Removal = 36 % Removal = 62 %
OCD Signal (au)

1 500 1 500

1 000 1 000
1

500 1 500
3

0 0
30 40 50 60 70 80 90 100 30 40 50 60 70 80 90 100

2 500 S = 0.1 c 2 500 S = 0.2 d

2 R = 10 R = 10
2 000 2 000
% COD % COD
Removal = 43 % Removal = 73 %
OCD Signal (au)

2
1 500 1 500

1 000 1 000
1
1
500 500
3 3
0 0
30 40 50 60 70 80 90 100 30 40 50 60 70 80 90 100

2 500 S = 0.1 e 2 500 S = 0.2 f

R=5 R=5
2 000 2 000
% COD % COD
Removal = 64 % Removal = 82 %
OCD Signal (au)

2 2
1 500 1 500

1 000 1 000

1
500 500
1
3 3
0 0
30 40 50 60 70 80 90 100 30 40 50 60 70 80 90 100

Retention Time (min) Retention Time (min)

Figure 5. Comparison of LC-OCD chromatograms for a subset of the AO treated samples for wastewater sample 1; the six panels are arranged according to
the S ¼ 0.1 (panels a, c, e) and S ¼ 0.2 (panels b, d, f) conditions in the DOE as originally shown in Figure 3a. The dashed line in each panel corresponds to
the control condition being pH adjustment of the as-received wastewater sample. The labels 1, 2, and 3 in each panel are used to identify peaks with
different retention times.

VOLUME 9999, 2017 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 7

 100 For the second section, it was found that the peak sizes changed
based on the specific treatment condition. For those conditions
90 that achieved COD removals less than 70 % (i.e. Figures 6a, 6b,
(Retention Time > 60 min) 6c, 6e), a ‘shoulder peak’ appeared on peak 1. The appearance of
% Reduction in peak area

80
this peak was somewhat unusual as it was not present in the
starting sample. In fact, this observation is supported by the other
five LC-OCD analyses for the centre point condition and the four
70 axial point conditions in the DOE method; refer to Figure S-2 in the
Supplementary Material. A previous study reported a similar
60 observation related to the treatment of surface water containing
NOM and attributed it to the decomposition of particulate
matter.[22] For those conditions that achieved COD removals
50
greater than 70 % (i.e. Figures 6d, 6f), it is hypothesized that a
two-stage process occurred during the AO treatment as follows:
40 first the particulate matter was decomposed (as mentioned above)
30 50 70 90 and then that material was further broken down. A schematic
% COD Removal summary of the proposed mechanism of COD removal in
wastewater 1 is provided in Figure 7a. Clearly, the integration
Figure 6. Comparison of analytical results from AO treatment of the of the LC-OCD analysis into the optimization of AO treatment
wastewater 1 sample; the dashed line is the linear regression model processes has yielded valuable information about the fate of
y ¼ 1.00x þ 17.88 (R2 ¼ 0.92). different species involved in the treatment process, however
further research is needed to validate the proposed mechanisms.
The LC-OCD chromatogram for each treated sample was divided As shown in Figure 8, the same presentation format was used
into two sections—peaks with retention times greater than 60 min for the LC-OCD chromatograms associated with the AO treatment
and those with retention times less than 60 min. of the wastewater 2 sample. Each panel corresponds to the same
For the first section, it was found that the peak size consistently specific S and R conditions as indicated in the panel legend; the
decreased across all treatment conditions. However, the magnitude six panels are arranged strategically to match the arrangement of
of the decrease in peak size varied across the different treatment the S ¼ 0.1 and S ¼ 0.2 conditions in the DOE setup that was
conditions. As shown in Figure 6, it was found that the magnitude originally presented in Figure 3. Again, it is interesting to note that
of this reduction (determined by LC-OCD analysis) was well these six conditions nearly span the full range of percent
correlated with the percent COD removal (determined via the COD removal that was found over the entire DOE method. In
HACH test kits). It should be noted that as the COD measurements the case of wastewater 2, formation of larger species caused by pH
were done on unfiltered samples, three different mechanisms adjustment is likely one of the factors resulting in the lower COD
contributed to the overall COD removals reported in this study: removal efficacy for this wastewater compared to wastewater
decomposition of particulate organics to dissolved organics; decom- 1 as larger species are potentially harder to decompose. This is
position of dissolved organics to products with higher oxidative in accordance with the findings of Lipczynska-Kochany and
levels compared to the original dissolved organics; complete Kochany[30] who reported that in the presence of humic
mineralization of the dissolved organics.[28] However, LC-OCD substances, acidic pH had a negative effect on the organic removal
cannot measure particulate organic material since all samples had of a coke plant wastewater heavily contaminated with ammonia.
to be filtered (0.45 mm filter) prior to LC-OCD measurement. Peak 3 represents dissolved organic matter of low molecular

a Particulate Organic Matter b Particulate Organic Matter

Formation of the Formation of peak 1
shoulder peak and and
enlargement of peaks enlargement of peak 2
1 and 2

Dissolved Organic Matter Dissolved Organic Matter

Shrinkage of the shoulder

peak, peak 1, and peak 2 Shrinkage of peak 2

Low Molecular Weight Dissolved Organic Low Molecular Weight Dissolved Organic
Matter Matter

Shrinkage of peaks Shrinkage of peak 3

with RT > 60 min

CO2 and H2O CO2 and H2O

Figure 7. Schematic illustration of the proposed mechanism of COD removal for both wastewater 1 (panel a) and wastewater 2 (panel b).

8 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 9999, 2017

 8 000 S = 0.1 a 8 000 S = 0.2 b
2
R = 15 R = 15
7 000 7 000
2
6 000 % COD 6 000 % COD
OCD Signal (au) Removal = 34 % Removal = 51 %
5 000 5 000

4 000 4 000

3 000 3 000
3 3
2 000 2 000 1
1
1 000 1 000

0 0
30 40 50 60 70 80 90 100 30 40 50 60 70 80 90 100

8 000 S = 0.1 c 8 000 S = 0.2 d

R = 10 R = 10
7 000 2 7 000

6 000 % COD 6 000 2 % COD

Removal = 37 % Removal = 54 %
OCD Signal (au)

5 000 5 000

4 000 4 000

3 000 3 000
3 3
2 000 2 000 1
1 000 1 1 000

0 0
30 40 50 60 70 80 90 100 30 40 50 60 70 80 90 100

8 000 S = 0.1 e 8 000 S = 0.2 f

R=5 R=5
7 000 7 000
2
6 000 % COD 6 000 % COD
Removal = 43 % 2 Removal = 53 %
OCD Signal (au)

5 000 5 000

4 000 4 000

3 000 3 000
3 3
2 000 2 000
1 1
1 000 1 000

0 0
30 40 50 60 70 80 90 100 30 40 50 60 70 80 90 100

Retention Time (min) Retention Time (min)

Figure 8. Comparison of LC-OCD chromatograms for a subset of the AO treated samples for wastewater sample 2; the six panels are arranged according to
the S ¼ 0.1 (panels a, c, e) and S ¼ 0.2 (panels b, d, f) conditions in the DOE as originally shown in Figure 3b. The dashed line in each panel corresponds to
the control condition involving just a pH adjustment of the as-received wastewater sample. The labels 1, 2, and 3 in each panel are used to identify peaks
with different retention times.

weight that was originally present in the as-received sample. At across the treatment conditions. As shown in Figure S-3 in
conditions with high COD removal this peak has been completely the Supplementary Material, a positive correlation was found
removed. A simple comparison to the chromatogram obtained between the percent COD removal and size of peak 1 while a
for the pH-adjusted sample clearly shows that a new peak, negative correlation was found between the percent COD
identified as peak 1 (at a retention time of approximately 46 min), removal and sizes of peaks 2 and 3. The observations presented
appeared in each LC-OCD chromatogram. Also, the extent of above for the wastewater 2 sample are in good agreement with the
decomposition of the species represented by peak 2 changed results of the regression analysis which indicated that COD

VOLUME 9999, 2017 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING 9

removal is a second order function of S, however, much less [4] P. V. Nidheesh, R. Gandhimathi, Desalination 2012, 299, 1.
significantly affected by R. A schematic summary of the proposed [5] A. Babuponnusami, K. Muthukumar, J. Environ. Chem. Eng.
mechanism of COD removal for wastewater 2 is provided in 2014, 2, 557.
Figure 7b. [6] J. Bautista, P. Mohedano, A. F. Zazo, J. Rodriguez, J. Chem.
Technol. Biot. 2008, 83, 1323.
CONCLUSIONS
[7] B. Bianco, I. De Michelis, F. Veglio , J. Hazard. Mater. 2011,
It was demonstrated that LC-OCD analysis complements DOE 186, 1733.
methods based on COD measurement for optimization of the AO
[8] N. San Sebastian Martınez, J. F. Fernandez, X. F. Segura,
process in the treatment of industrial wastewater with dynami-
A. S. Ferrer, J. Hazard. Mater. 2003, 101, 315.
cally varying composition. LC-OCD provides valuable information
about the molecular weight distribution of the organic pollutants [9] I. Arslan-Alaton, S. Dogruel, J. Hazard. Mater. 2004, 112,
in the wastewater and the alteration of these organics in the 105.
different steps of treatment. This information can be utilized to [10] F. Torrades, S. Saiz, J. A. Garcıa-Hortal, Desalination 2011,
better understand the effect of different treatment conditions on 268, 97.
the different COD removal steps and also to study the dependence [11] M. Ahmadi, F. Vahabzadeh, B. Bonakdarpour, E. Mofarrah,
of optimum treatment conditions on the wastewater composition. M. Mehranian, J. Hazard. Mater. 2005, 123, 187.
It is anticipated that LC-OCD analysis would complement [12] M. Pérez, F. Torrades, X. Domenech, J. Peral, Water Res.
DOE methods based on other bulk measurements (i.e. TOC and 2002, 36, 2703.
BOD) as well.  ski, K. Filipek, Polish J. Environ. Stud. 2001, 10,
[13] K. Barbusin
NOMENCLATURE 207.
[14] M. A. Oturan, N. Oturan, M. C. Edelahi, F. I. Podvorica, K. El
AO advanced oxidation
Kacemi, Chem. Eng. J. 2011, 171, 127.
BOD biochemical oxygen demand (mg/L)
[15] R. Salazar, S. Garcia-Segura, M. S. Ureta-Zan ~ artu, E. Brillas,
CCD central composite design
CI confidence interval Electrochim. Acta 2011, 56, 6371.
COD chemical oxygen demand (mg/L) [16] S. H. Yuan, X. H. Lu, J. Hazard. Mater. 2005, 118, 85.
DOC dissolved organic carbon (mg/L) [17] S. A. Huber, A. Balz, M. Abert, W. Pronk, Water Res. 2011,
DOE design of experiments 45, 879.
GC-MS gas chromatography-mass spectrometry [18] D. Metcalfe, P. Jarvis, C. Rockey, S. Judd, Sep. Purif. Technol.
HPLC high performance liquid chromatography 2016, 163, 173.
LC-MS liquid chromatography-mass spectrometry [19] X. Zheng, R. Mehrez, M. Jekel, M. Ernst, Desalination 2009,
LC-OCD liquid chromatography-organic carbon detection 249, 591.
NOM natural organic matter
[20] S. Jeong, Y. Okour, T. V. Nguyen, H. K. Shon, S.
OCD organic carbon detector
Vigneswaran, Desalin. Water Treat. 2013, 51, 3591.
OND organic nitrogen detector
R ratio of hydrogen peroxide molar concentration over [21] S. Ciputra, A. Antony, R. Phillips, D. Richardson, G. Leslie,
ferrous ion molar concentration Chemosphere 2010, 81, 86.
RT retention time (min) [22] G. O. Bosire, J. C. Ngila, Anal. Methods-UK 2016, 8, 1415.
S ratio of hydrogen peroxide molar concentration over [23] F. Chen, S. Peldszus, A. M. Elhadidy, R. L. Legge, M. I. Van
the COD of the wastewater sample ((mol/m3)/(mg/L)) Dyke, P. M. Huck, Water Res. 2016, 104, 361.
TOC total organic carbon (mg/L) [24] A. Sebastian Ruhl, M. Jekel, J. Water Supply Res. T. 2012,
UV ultraviolet 61, 32.
UVD ultraviolet absorption detector
[25] R Core Team, R: A language and environment for statistical
computing, R Foundation for Statistical Computing, Vienna
ACKNOWLEDGEMENTS 2016, http://www.R-project.org/
[26] F. Torrades, S. Saiz, J. A. Garcıa-Hortal, Desalination 2011,
The authors acknowledge Aevitas for the funding support for
268, 97.
the project and for providing the wastewater samples. Also, we
thank Rachel Trower and Monica Tudorancea at the University of [27] H. Zhang, H. J. Choi, C.-P. Huang, J. Hazard. Mater. 2005,
Waterloo for their assistance with the LC-OCD instrument 125, 166.
calibration and data analysis. [28] A. E. Papadopoulos, D. Fatta, M. Loizidou, J. Hazard. Mater.
2007, 146, 558.
REFERENCES [29] G. Yohannes, S. K. Wiedmer, M. Jussila, M.-L. Riekkola,
Chromatographia 2005, 61, 359.
[1] I. Oller, S. Malato, J. A. Sanchez-Pérez, Sci. Total Environ.
2011, 409, 4166. [30] E. Lipczynska-Kochany, J. Kochany, Chemosphere 2008,
73, 745.
[2] M. A. Oturan, J.-J. Aaron, Crit. Rev. Environ. Sci. Tec. 2014,
44, 2641.
[3] I. N. Dias, B. S. Souza, J. H. O. S. Pereira, F. C. Moreira, M. Manuscript received December 23, 2016; revised manuscript
Dezotti, R. A. R. Boaventura, V. J. P. Vilar, Chem. Eng. J. received February 23, 2017; accepted for publication March 22,
2014, 247, 302. 2017.

10 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING VOLUME 9999, 2017