Abstract

We improved the ultraviolet (UV)/O3-based method for measuring chemical oxygen demand (COD) in water. An on-line COD monitoring device was developed and the UV/O3 method was used to oxidize sample solutions. A model was established by using support vector machines (SVM) algorithm to estimate dissolved oxygen and CO2 in solutions. Based on the measured data by each sensor during the oxidation process and the estimated dissolved oxygen and CO2, the UV/O3-based COD test accuracy was improved. This approach overcomes many problems associated with the conventional COD determination techniques such as long analysis time, consumption of expensive and toxic reagents, and production of secondary toxic waste. The effect of important parameters on the measurement of COD was systematically investigated. The improved method was successfully applied to determine the COD of real samples from environmental water. Compared with the standard dichromate method, our UV/O3-based COD test method is more effective. The assay time of 10–15 min/sample can be readily achieved. A practical detection limit of 0.89 mg/L COD with a linear range of 1–300 mg/L was achieved under the normal conditions.

INTRODUCTION

The per capita water resource in China is only a quarter of the world average. With the rapid development of economy, the growing problem of water pollution in China, including water pollution coming from industrial wastewater and domestic sewage, has greatly impacted human health, which results in irreversible economic losses and a potential threat for the sustainable development of China's economy (Javier et al. 2015; Jiang 2015; Wang & Yang 2016). Therefore, water quality monitoring and water pollution control are essential for China.

Chemical oxygen demand (COD) test is commonly used to monitor water quality (Mamais et al. 1993). The test measures the mass of oxygen consumed to oxidize the organic pollutants in polluted water, with the unit of milligrams of oxygen per liter (mg/L) (Singh et al. 2013). Therefore, COD reflects the amount of organic pollutants in water, an important parameter to control the quality of discharged water (Wu et al. 2014).

Common COD test methods include the dichromate oxidation method (CODCr) (Dedkov et al. 2000), the permanganate index method (CODMn) (Tsuboi et al. 2004; Zenki et al. 2006), the ultraviolet (UV) spectrophotometry method (Brookman 1997), the microwave digestion method (Jardim & Rohwedder 1989), and the electrochemical method (Moussavi & Aghanejad 2014). The CODCr method is a reliable national standard method, but the process is long and needs several chemical reagents, which increases operating costs (Kawai et al. 2015; Darajeh et al. 2016). The method can also cause secondary pollution. Other methods also have a series of problems such as complicated operation, unstable detection, and high maintenance cost (Yang et al. 2014). Apart from these methods, the UV/O3 treatment is one of the most effective and environment-friendly methods to oxidize organic species (Fu et al. 2016; Stapf et al. 2016). Zhang & Yang (2014) used the UV/O3 method to oxidize organic pollutants in water samples, and measured the amount of O3 and O2 at the inlet and outlet of the reactor by sensors. The COD value was estimated based on the amount of O3 consumption and O2 increment in the process of COD test. This method reduced the test time, did not use chemical reagents, and thus avoided secondary pollution. However, this method did not take into account the impact of CO2 generated during the oxidation of organic pollutants, which increases the uncertainty of measurement error. Based on the UV/O3 method, Zhang et al. (2015) developed a COD calculation model by taking the CO2 generated from the oxidation process into account. However, the test results were still inaccurate until compensated by an exponential model. The compensation model did not consider the effect of dissolved gases after the reaction. Our experiments showed that a certain amount of O2, O3 and CO2 would dissolve in solutions after the reaction, which affected the COD value. If we ignore the dissolved gases, the COD test error would increase. Therefore, it is necessary to consider the amount of dissolved gases in the COD calculation model.

At present, our device is a prototype, so the instrumentation used for the measurement of COD is more complex. In order to ensure the accuracy of the measurement, we used a number of sensors with high sensitivity and accuracy which sacrificed the range of the sensor. So our device may not be suitable for measuring high COD concentration of industrial wastewater, but suitable for the detection of environmental water.

In this paper, we experimentally measured the dissolved oxygen and CO2 in the sample solutions based on Zhang et al. (2015). Experimental data such as O3 consumption, CO2 production, oxidation temperature and pressure were measured by various sensors. A model was developed to estimate dissolved oxygen and CO2 based on a support vector machine leaning algorithm (Fischetti 2016). In addition, we modified the COD calculation model to improve its accuracy.

EXPERIMENTS

COD measurement system

A schematic diagram of the UV/O3 COD measurement system is shown in Figure 1. The system included an ozone generator (DJ800, Jinhua Guangyuan Instrument Factory), a temperature sensor (DS18b20, Dallas Co., Ltd), a pressure sensor (YB-131, Shanghai Zhengbao pressure meter), a quartz reactor, two UV lamps (ZW28D15 W(Y)-Z550, Guangdong Cnlight Co., Ltd), a CO2 concentration meter (FS4001, Siargo Co., Ltd), two ozone concentration meters (JSK-A2, Shenzhen Jin Shikai Technology Co., Ltd), two gas flow sensors (FS4001, Siargo Co., Ltd), a data collector, and a computer. Except for the inlet and outlet, the whole system was airproof. Experiments of sample solutions oxidation were conducted using a semi-batch reactor made of quartz. The reactor was a vertically oriented tube with a length of 60 cm and an inner diameter of 1.5 cm. During experiments, the tube was filled with 15 mL solution. Two UV lamps, with a primary wavelength of 254 nm, were placed on both sides of the reactor. Spacing between the quartz tube and UV lamps was kept at 1 cm. Ozone gas was generated by electrolysis of water and bubbled into the reactor and exhaust ozone was released through the outlet. During the oxidation process, concentration of O3 and CO2, gas flow rate, temperature, and pressure were monitored by sensors.

Figure 1

Schematic diagram of the UV/O3 COD measurement system.

Figure 1

Schematic diagram of the UV/O3 COD measurement system.

Solutions preparation

The potassium acid phthalate (C8H5KO4, FW 204.22), glucose (C6H12O6, FW 180.16), potassium iodide (KI, FW 166.00), and potassium permanganate (KMnO4, FW 158.03) were analytical grade (Sinopharm Chemical Reagent Co., Ltd). Deionized water (DI water, 15–18.2 MΩ•cm) produced from a Hitech Smart-Q15 system (Hitech Instruments Co., Ltd, Shanghai) was used as solvent.

The standard COD stock solution was prepared by dissolving 0.4251 g C8H5KO4 into 500 mL of DI water. This corresponded to a COD concentration of 1,000 mg/L. By diluting the stock solution, standard solutions with COD concentrations of 20, 40, 60, 80 and 100 mg/L COD were obtained. Glucose solutions with equivalent COD concentrations ranging from 20 to 100 mg/L were also prepared.

COD calculation

In UV/O3 treatment, oxidation of organic pollutants is mainly caused by hydroxyl radical (OH•) generated by UV photolysis of O3. The chemical process can be summarized as the following reactions (Gurol & Aysegul 1996).  
formula
(1)
Assuming the reaction time is t seconds and the sampling period during the oxidation process is 1 second, the O3 concentration at inlet and outlet at the ith second can be recorded as and , respectively. The CO2 concentration at outlet can be recorded as . The gas flow at inlet and outlet can be recorded as and , respectively. While the temperature and pressure at outlet can be recorded as and .
According to the ideal gas state formula of , the gas flow sensor was calibrated under the standard conditions (, ) using the O2 as the medium (the densities of O2 and O3 are and in this state). The O3 generated by the ozone generator is composed of O3 and O2 with a mass ratio of 1: 4, and the density of the mixed gas is obtained as follows.  
formula
(2)
The gas flow measured under operating conditions should be converted to the standard conditions. The converted gas flow was denoted as and . Since the mass of the gas was constant, Equations (3) and (4) were obtained after conversion.  
formula
(3)
 
formula
(4)
The inlet gas is mainly a mixture of O3 and O2 generated by the ozone generator. As the oxidation process continues, the organic pollutants is oxidized to generate CO2, so the outlet gas is mainly O3, O2 and CO2. Therefore, according to the compositions of mixed gases at inlet and outlet and parameters measured by sensors, the amounts of O3 (, ), CO2 and O2 (, ) at the ith second can be calculated as follows.  
formula
(5)
 
formula
(6)
 
formula
(7)
 
formula
(8)
 
formula
(9)
In order to calculate the consumed O3 during the oxidation process, the difference between the instantaneous O3 concentrations at the inlet and outlet is integrated with the reaction time. In the same way, the O2 increment can be obtained. So the O3 consumption and O2 increment can be expressed as follows.  
formula
(10)
 
formula
(11)
According to the reaction principle, the generated free oxygen radical (O•) can be obtained based on O3 consumption and O2 increment. The O• can rapidly react with water to generate OH• through the free radical reaction (see reactions (1)). According to the second reaction in the system (1), the ratio between mass of OH• generated and O• produced is 34/16. Because the actual oxidant for the oxidation is OH•, the COD can be calculated with the following equation.  
formula
(12)
where V was the volume of the sample solution, x is the amount of generated O•, and y is the amount of generated OH•.
During the oxidation process, part of O2, O3 and CO2 can dissolve in sample solutions. The O3 consumption and CO2 production can be rewritten as follows.  
formula
(13)
 
formula
(14)
where B and C are dissolved oxygen and CO2, respectively. So the cumulative value of the O2 at the outlet can be rewritten as:  
formula
(15)
The O2 increment can be rewritten as:  
formula
(16)
Finally, the COD calculation formula can be modified as shown in Equation (17).  
formula
(17)

Experimental procedures

The organic species was oxidized under ambient temperature and pressure with the UV/O3 treatments. Before each experiment, ozone generator, UV lamps, and sensors were warmed up to achieve stability. The reactor was cleaned twice with deionized (DI) water every time before injecting a new sample solution. For each experiment, 15 mL of sample solution was injected into the reactor. The O3 was stably bubbled into the reactor surrounded by UV irradiation until the oxidation process finished. During the process, data were automatically collected by multiple sensors through a LabView program and stored in the computer. All samples after oxidation were drained into a beaker and settled for 1 min to remove the O3 gas above liquid level. Manganese sulfate and alkaline potassium iodide were then mixed with the sample for a full reaction. The amount of dissolved oxygen was determined by titrimetry using starch as the indicator. The amount of dissolved CO2 was estimated by comparing the theoretical and actual amount of CO2 production. With the help of support vector machine, a model was established to estimate the amount of dissolved oxygen and CO2. The COD value was finally calculated with Equation (17).

RESULTS AND DISCUSSION

Estimation of dissolved CO2 in solution

Standard solutions with different concentrations of organic chemicals and thus different COD values were still prepared by C8H5KO4. After oxidation, the amounts of CO2 produced in each standard solution were compared with the theoretical values as shown in Figure 2.

Figure 2

Comparison of CO2 determination with theoretical value.

Figure 2

Comparison of CO2 determination with theoretical value.

From Figure 2, the measured CO2 values were less than the theoretical values calculated from the chemical equation. Since CO2 is a relatively stable gas and cannot be oxidized by O3 or O2 under UV light, the measured CO2 values were less than the theoretical values probably because some CO2 dissolved into the sample solutions. The CO2 in water mainly took the forms of carbon dioxide molecules , carbonic acid , carbonate and bicarbonate . The percentage and content of each form are complex (Søndergaard & Schierup 1982). Therefore, it was difficult to measure the dissolved carbon dioxide in water directly. The difference between the theoretical and measured CO2 values was used to compensate the measured values to get a better estimation.

Model to estimate dissolved oxygen and CO2

First of all, COD standard solutions were prepared with potassium acid phthalate and glucose, each solution had 60 groups (0–100 mg/L, concentration difference of 20 mg/L, each concentration of 10 groups). Then, these sample solutions were oxidized by the UV/O3 COD measuring equipment. Finally, experimental data of 120 groups of standard solutions were obtained after reaction. The amount of dissolved gas was related to the COD concentration and the amount of gas generated. Generally, higher gas concentration, lower temperature, and higher pressure can cause a higher amount of dissolved gas. The data obtained by sensors during the oxidation process, such as O3 consumption, CO2 production, oxidation temperature and pressure, were used to establish a soft-sensing model to estimate the dissolved O3 and CO2. Since the support vector machine (SVM) is perfect for leaning based on small sample size, a model was established to estimate dissolved oxygen and CO2 by using SVM.

The O3 consumption (), CO2 production (), oxidation temperature () and pressure () during the oxidation process were selected as input variables of the estimation model. The output variables of the model were dissolved oxygen () and dissolved CO2 (). In this paper, the Gaussian radial basis kernel function was chosen as the kernel function of SVM. The nuclear bandwidth () and penalty factor () can affect the precision of the model. Through plenty of experiments, we got the values of and . Among all the 120 groups of experimental data, 80 groups were selected as the training set of the estimation model (), and the remaining 40 groups of data were used as the test set of the model. The estimated values of dissolved oxygen and CO2 with the model were shown in Figures 3 and 4. From these figures, the estimated values of the test set were close to the real values, with the maximum relative errors being 4.63% and 4.85%, respectively. The model also accurately estimated the fluctuation between the theoretical and estimated values, and thus can replace manual tests to determine the dissolved oxygen and CO2.

Figure 3

Estimated results of dissolved oxygen.

Figure 3

Estimated results of dissolved oxygen.

Figure 4

Estimated results of dissolved carbon dioxide.

Figure 4

Estimated results of dissolved carbon dioxide.

Effect of dissolved oxygen and CO2

In order to study the effects of dissolved oxygen and CO2 on the results of COD determination, mixed solutions of potassium acid phthalate and glucose with the mass ratio of 1:1 were prepared. The concentrations of COD solutions were 20, 40, 60, 80 and 100 mg/L, respectively. The calculated COD values according to Equations (12) and (17) were taken as the original values and modified values respectively. Furthermore, values of average COD, relative standard deviation (RSD) and mean relative error (MRE) were calculated and listed in Tables 1 and 2. The values of average COD was obtained by averaging the data of ten sets of COD values for each sample. The RSD and MRE were computed using Equations (18) and (19). The defines the standard deviation of measured COD values in the same sample. Here, stands for the known concentration of standard COD sample solutions prepared. The results in Table 2 were closer to the standard value than the results in Table 1. Although the RSD values were similar, the MRE values in Table 2 were greatly reduced. It revealed that the improved UV/O3 method for COD measurement was more accurate and stable at the lower concentration end.  
formula
(18)
 
formula
(19)
Table 1

Analysis of original measured COD value

Standard COD samples (mg/L) Average value of measured COD (mg/L) RSD (%) MRE (%) 
20 23.69 4.03 18.45 
40 43.93 2.5 9.82 
60 57.49 1.81 4.17 
80 76.44 1.19 4.45 
100 93.54 1.21 6.46 
Standard COD samples (mg/L) Average value of measured COD (mg/L) RSD (%) MRE (%) 
20 23.69 4.03 18.45 
40 43.93 2.5 9.82 
60 57.49 1.81 4.17 
80 76.44 1.19 4.45 
100 93.54 1.21 6.46 
Table 2

Analysis of modified measured COD value

Standard COD samples (mg/L) Average value of measured COD (mg/L) RSD (%) MRE (%) 
20 21.09 3.59 5.43 
40 41.86 1.88 4.65 
60 58.76 1.23 2.06 
80 81.77 1.11 2.21 
100 103.35 1.04 3.34 
Standard COD samples (mg/L) Average value of measured COD (mg/L) RSD (%) MRE (%) 
20 21.09 3.59 5.43 
40 41.86 1.88 4.65 
60 58.76 1.23 2.06 
80 81.77 1.11 2.21 
100 103.35 1.04 3.34 

The CODCr method was used to validate the improved UV/O3 method for COD determination. The COD standard solutions were still prepared with potassium acid phthalate and glucose and their COD values were measured with both the CODCr method and the improved UV/O3 method (Table 3). From Table 3, the MRE of the two methods was less than 5%, which examined the applicability of the proposed detection principle by using the synthetic samples.

Table 3

Comparison of methods for COD determination

Improved UV/O3 method (mg/L) National standard dichromate method (mg/L) MRE (%) 
21.09 20.4 3.38 
41.86 40.3 3.87 
58.76 59.5 1.24 
81.77 79.4 2.98 
103.35 98.7 4.71 
Improved UV/O3 method (mg/L) National standard dichromate method (mg/L) MRE (%) 
21.09 20.4 3.38 
41.86 40.3 3.87 
58.76 59.5 1.24 
81.77 79.4 2.98 
103.35 98.7 4.71 

Analysis and assessment of gas flow

As the improved UV/O3 method derives its information from several sensors, there are a number of variables that need to be investigated in terms of how they affect the accuracy of the COD measurements when used for on-line, real-time monitoring. Through a large number of repetitive experiments, it was found that the gas flow is the key factor which affects the accuracy of COD measurements directly or indirectly.

When the electrolysis-type ozone generator was used for a long time, the exchange membrane in it would gradually dissipate, resulting in the continued reduction of the gas produced. Figure 5 shows the curve of gas flow with time of use. After the cumulative use of 200 h, a direct proportional correlation can be established with (where is the gas flow at inlet and t is the accumulated working hours). The correlation coefficient R2 for the linear fit is 0.982 in the time range of 200–350 h.

Figure 5

Curve of gas flow with time of use.

Figure 5

Curve of gas flow with time of use.

As the gas flow continued to change, according to Equations (5)–(7), the amounts of O3 and CO2 would be affected directly. So the calculation results such as O3 consumption and CO2 production would be affected as well. For a given sample with known COD concentration of 20 mg/L, the change of O3 consumption and CO2 production by gas flow was shown in Figure 6, where all experiments were carried out under identical conditions except gas flow. It was found that a decrease of gas flow results in a decrease of O3 consumption and CO2 production. This is because a slow flow rate allows longer contact time for the CO2 to dissolve in water. However, the effect of gas flow on O3 consumption is relatively small for the reason that O3 is easier to saturate.

Figure 6

Effect of gas flow on O3 consumption and CO2 production.

Figure 6

Effect of gas flow on O3 consumption and CO2 production.

More importantly, the gas flow can influence the detection sensitivity and linear range indirectly. Table 4 shows the effect of gas flow on the detection limits and linear range. It was found that when gas flow was decreased from 152 to 122 sccm, the detection limit was changed from 0.89 to 1.06 mg/L. In addition to the reduction in detection limits (sensitivity), the reduction of gas flow can reduce the liner range.

Table 4

Effect of gas flow on the measurement of COD on detection limit and linear range

Gas flow (sccm) Detection limit (mg/L) Linear range (mg/L) 
152 0.89 1–300 
146 0.92 1–290 
141 0.95 1.5–280 
136 0.98 1.5–270 
128 1.02 1.5–260 
122 1.06 1.5–250 
Gas flow (sccm) Detection limit (mg/L) Linear range (mg/L) 
152 0.89 1–300 
146 0.92 1–290 
141 0.95 1.5–280 
136 0.98 1.5–270 
128 1.02 1.5–260 
122 1.06 1.5–250 

According to the definition from International Union of Pure and Applied Chemistry (IUPAC), COD detection limit can be determined by three times of the standard deviation divided by the slope of calibration curve. So the current method has a COD detection limit of 0.89 mg/L with a working range for COD concentration between 1 and 300 mg/L. Additionally, the regular replacement of the ozone generator can help improve the stability of the device.

Real sample analysis

The applicability of the improved UV/O3 method for real sample analysis was examined. A total of 10 water samples were collected from different rivers in the city for examination. Each sample was analyzed by both the improved UV/O3 method and the standard dichromate method. Figure 7 shows the correlation between the COD values obtained by both methods.

Figure 7

Correlation between the improved UV/O3 method and the standard dichromate method for the real water samples.

Figure 7

Correlation between the improved UV/O3 method and the standard dichromate method for the real water samples.

The Pearson correlation coefficient (Zhang et al. 2006) was used as a measure of the intensity of association between the two methods. A highly significant correlation (r = 0.992) between the two methods was obtained, indicating the two methods agreed very well. And the slope of the principal axis of the correlation ellipse of 1.036 was obtained. The almost identical slope values suggest both methods were accurately measuring the same COD value. Given a 95% confidence interval, this slope was between 1.026 and 1.046. This implies that we can be 95% confident that the true slope lies between these two values. The strong correlation and almost unity in slope obtained demonstrate the applicability of improved UV/O3 method for determination of COD in environmental water.

CONCLUSIONS

We improved the UV/O3-based method for measuring COD in polluted water and developed a COD on-line detection device. A model was established to estimate dissolved oxygen and CO2 by using SVM. The experimental results showed that the model was able to estimate dissolved oxygen and CO2 in sample solutions according to the O3 consumption, the CO2 production, and the oxidation temperature and pressure. The improved UV/O3-based method could measure the COD with accuracy similar to the national standard method. The method we developed could be used for rapid on-line detection of COD.

ACKNOWLEDGEMENTS

This work was supported by the Fundamental Research Funds for the Central Universities (JUSRP51733B), and the National Natural Science Foundation of China (61773181).

REFERENCES

REFERENCES
Darajeh
,
N.
,
Idris
,
A.
,
Fard Masoumi
,
H. R.
,
Nourani
,
A.
,
Truong
,
P.
&
Sairi
,
N. A.
2016
Modeling BOD and COD removal from palm oil mill secondary effluent in floating wetland by Chrysopogon zizanioides (L.) using response surface methodology
.
Journal of Environmental Management
181
,
343
352
.
Dedkov
,
Y. M.
,
Elizarova
,
O. V.
&
Kel'Ina
,
S. Y.
2000
Dichromate method for the determination of chemical oxygen demand
.
Journal of Analytical Chemistry
55
(
8
),
777
781
.
Fu
,
P.
,
Feng
,
J.
,
Yang
,
H.
&
Yang
,
T.
2016
Degradation of sodium n-butyl xanthate by vacuum UV-ozone (VUV/O3) in comparison with ozone and VUV photolysis
.
Process Safety & Environmental Protection
102
,
64
70
.
Gurol
,
M. D.
&
Aysegul
,
A.
1996
Kinetics of ozone photolysis in aqueous solution
.
Aiche Journal
42
(
11
),
3283
3292
.
Jardim
,
W. F.
&
Rohwedder
,
J. J. R.
1989
Chemical oxygen demand (COD) using microwave digestion
.
Water Research
23
(
8
),
1069
1071
.
Javier
,
G. P.
,
Virginia
,
L.
,
Gonzalo
,
L. A.
,
Mario
,
G. S.
&
Pablo
,
F. N.
2015
Ovarian cancer mortality and industrial pollution
.
Environmental Pollution
205
,
103
110
.
Kawai
,
M.
,
Nagao
,
N.
,
Kawasaki
,
N.
,
Imai
,
A.
&
Toda
,
T.
2015
Improvement of cod removal by controlling the substrate degradability during the anaerobic digestion of recalcitrant wastewater
.
Journal of Environmental Management
181
,
838
846
.
Singh
,
S. K.
,
Tang
,
W. Z.
&
Tachiev
,
G.
2013
Fenton treatment of landfill leachate under different cod loading factors
.
Waste Management
33
(
10
),
2116
2122
.
Søndergaard
,
M.
&
Schierup
,
H. H.
1982
Dissolved organic carbon during a spring diatom bloom in lake Mossø, Denmark
.
Water Research
16
(
6
),
815
821
.
Wu
,
J.
,
Yan
,
G.
,
Zhou
,
G.
&
Xu
,
T.
2014
Wastewater cod biodegradability fractionated by simple physical–chemical analysis
.
Chemical Engineering Journal
258
,
450
459
.
Yang
,
S.
,
Liang
,
Z.
,
Yu
,
H.
,
Li
,
X.
,
Lu
,
K.
&
Chen
,
Y.
2014
Study on COD removal mechanism of aminosilicone polymer microemulsion wastewater under fenton oxidation conditions
.
Desalination & Water Treatment
52
(
34–36
),
6777
6783
.
Zhang
,
L.
&
Yang
,
H. Z.
2014
COD detection device based on ozone coupled with UV
.
China Water & Wastewater
30
(
6
),
89
91
.
Zhang
,
L.
,
Xue
,
P.
&
Yang
,
H. Z.
2015
Study of on-line detection of cod based on ozone/UV
.
Chinese Journal of Environmental Engineering
9
(
1
),
235
239
.