Source Identification of VOCs in a Petrochemical Complex by Applying Open-Path Fourier Transform Infrared Spectrometry

The petrochemical complex examined in this work includes a great variety of facilities and factories that emit various odorants. A stationary open-path Fourier transform infrared (OPFTIR) system can be used for routine VOC and odor monitoring. However, when odor episodes occur, only multiple mobile OPFTIR systems are able to identify the odorant sources effectively and efficiently. In this study, N,N-dimethyl formamide (DMF) was found to be the most commonly detected odorant emitted from the investigated petrochemical complex by routine monitoring of a stationary OPFTIR system. Then the sequential deployments of a pair of mobile OPFTIR systems were carried out both upwind and downwind of the different sections of the focal area. The pollution rose plots derived from the data obtained by the pair of mobile OPFTIR systems identified the DMF sources. By conducting correlation analyses on the data obtained from the mobile OPFTIR situated at the downwind location of the DMF emission sources, we found that besides DMF, the dry PU synthetic leather process of plant B was also the major odorant source of 2-butanone, ethyl acetate and isopropanol. The source identification measure developed in this study can be used to clarify possible odorant sources not only for petrochemical industrial complexes but also for other areas associated with various emission sources.


INTRODUCTION
The operation of petrochemical plants leads to the emission of various air pollutants, such as highly reactive hydrocarbons (Ryerson et al., 2003), polycyclic aromatic hydrocarbons (PAHs) (Wang et al., 2002;Wang et al., 2007;Wang et al., 2009), volatile organic compounds (VOCs) (Cetin et al., 2003) and odors (Chen et al., 2000;Dincer and Muezzinoglu, 2006) that have significant impacts on ambient air quality and adverse effects on human health (Chang et al., 2013).One issue often reported by local residents is that of unpleasant odors, and in practice the air pollution problems related to such facilities are mostly recognized as odor problems by the public (Dincer and Muezzinoglu, 2006).
The characterization and measurement of odorants is a challenging but important environmental issue.The odor problems related to petrochemical plants are often caused by VOCs emitted from stacks, leakages from pipelines and tanks, production processes and wastewater treatment (Kalabokas et al., 2001;Huang et al., 2012;Chen et al., 2013).People living near a petrochemical complex can thus suffer from direct and indirect exposure to unpleasant odors.If the odor sources cannot be located and controlled, then this could lead to tensions among the government, local residents, and the operators of the complex.However, sometimes it is very difficult to find the sources of unpleasant odors, as petrochemical complexes include a great variety of facilities and factories that could emit VOCs.
Although GC/MS spectrometry can determine the specific compounds of VOCs, the samples need to be collected and sent back to laboratories for chemical analysis (Ho et al., 2013;Lai et al., 2013).Furthermore, using single point monitoring to profile a large area is not a cost-effective sampling strategy, and is not able to provide the real-time measurements which are crucial for monitoring the inherently intermittent odor episodes.In contrast, open-path Fourier transform infrared (FTIR) spectrometry has the advantages of being adaptable, transportable, capable of remotely monitoring several different paths in real time, and carrying out multi-compound analysis.This approach has thus been widely applied for air pollutant monitoring and odor source identification at large-area sources, such as semiconductor manufacturing plants (Tsao et al., 2011), swine production facilities (Childers et al., 2001;Aneja et al., 2008), coke ovens (Lin et al., 2007), petrochemical plants (David et al., 1994;Webb et al., 1996), and urban areas (Hong et al., 2004).Several studies also showed that the concentrations of target compounds obtained by OPFTIR are comparable to those determined by conventional methods (Childers et al., 2001).
Compared to the odor sources in previous study which applied OPFTIR for source identification, a petrochemical complex has far more manufacturing processes and facilities that could emit concerned odorants and fugitive VOCs.Therefore, the objective of this study was to develop an integrated approach to successful monitor and identify the sources of odorants emitted from a petrochemical industrial complex by applying stationary and multiple mobile OPFTIR systems.The sources of the odorants and the fugitive VOC characteristics were also investigated to evaluate the effectiveness and feasibility of using OPFTIR systems to monitor a petrochemical complex, where many manufacturing processes and facilities are densely situated.

The Investigated Petrochemical Complex
A total of fifteen petrochemical plants are situated at the investigated petrochemical complex, and each plant had its own stacks, pipelines, tanks, production processes, and wastewater treatment facilities that can emit odorants.The location of the stationary OPFTIR system is located in the northwest side of the petrochemical complex, as shown in Fig. 1.

Description of the OPFTIR Systems
The stationary and mobile OPFTIR systems utilized in this study were manufactured by Mastek Technologies, Inc. (Taiwan).Each OPFTIR system consists of an FTIR unit, a retroreflector, and an operational computer.Each FTIR unit is equipped with a mid-band mercury cadmium telluride (MCT) detector, cooled by a liquid nitrogen automatic filling system to maintain the MCT detector below -198°C.Each retroreflector consists of a corner cube protected by a gold surface, and is mounted on a tripod at the opposite end of the monitoring path.The OPFTIR system is operated by EEL-IRSoft software, which was developed and authorized by the Taiwan Industrial Technology Research Institute, and supplied by Mastek Technologies, Inc.
The operations of the OPFTIR systems and data analyses followed U.S. EPA Method TO-16 and Taiwan EPA Method NIEA A002.10C.The FTIR data were further analyzed by using a multilevel classical least-square (CLS) method.Each spectral data consisted of 64 co-added interferograms, which were recorded every five minutes from 500 cm -1 to 4500 cm -1 at a nominal 1 cm -1 resolution.Furthermore, a mobile meteorological observation system was deployed to record wind speed and direction, as well as temperature, solar irradiance and humidity, to help identify the sources of odorants.
The detection limit of various VOCs, the coefficient of variation and accuracy of the mobile OPFTIR are listed in Tables S1 and S2.The coefficient of variation of the mobile OPFTIR for five VOC compounds were between 0.73%-5.8%,while the accuracy were between 89.6%-111%.

Strategy of Deploying OPFTIR Systems
The stationary OPFTIR system was set up for routine VOC monitoring, and unable to identify the sources of VOCs.When the incidence rate of odor complaints increased or elevated atmospheric VOC concentrations were detected by the stationary OPFTIR system, the pair of mobile OPFTIR systems was then deployed in both upwind and downwind of a section of the petrochemical complex according to the meteorological conditions.After sequentially deploying the pair of mobile OPFTIR systems on the different sections, the major odorants and their sources of the petrochemical complex were characterized and identified.The pollution rose plots derived from the data from the pair of mobile OPFTIR systems were then used to judge the locations of the odorant sources.The detailed path lengths and heights of the OPFTIR settings are listed in Table S3.

Regular Monitoring by the Stationary OPFTIR System
After a half year's continuous monitoring, 24 VOCs were detected by the stationary OPFTIR system, including N,N-dimethyl formamide (DMF), vinyl acetate, ethyl acetate, 1,3-butadiene, 2-butanone, ethylene, propylene, cyclohexane, styrene, toluene, 1,1, 1,2-tetrafluoroethane, acetylene, ammonia, butane, bromotrifluoromethane, chlorodifluoromethane, chloroform, dichlorodifluoromethane, difluoromethane, methanol, nonanol, n-octane, pentafluoroethane, and sulfur hexafluoride.Among these compounds, DMF, which is often used as solvent for chemical reactions, as well as in the manufacturing of adhesives, synthetic leathers, and surface coatings, was the most commonly detected odorant that exceeded the air quality regulation of 200 ppb and the odor threshold of 470 ppb.DMF has been linked to cancer in humans, and according to IARC, DMF is a possible carcinogen.The temporal trends of DMF detected by the stationary OPFTIR system are shown in Fig. 2. As the prevailing wind direction gradually changed from the northeast in January to the southeast in June, the detection frequency (the detectable data points divided by total effective data points) of DMF by the stationary OPFTIR system located in downwind direction increased significantly, from below 0.5% to about 16%.Furthermore, the percentages by which the DMF concentrations exceeded the air quality regulation and odor threshold value also increased 33 times (from 0.2% to 5.3%) and 18 times (from 0.07% to 1.3%), respectively.
After a detailed survey of the raw materials, production processes, intermediate products and emissions that are related to DMF, based on information obtained from the EPA's management database, only three plants at the complex were found out to use DMF as a raw material (labeled as A, B and C in Fig. 1).The pollution rose plot of DMF (shown in Fig. 1) shows that the increase in DMF concentrations and detection frequency occurred when the southeastern winds prevailed, and thus the most likely  sources of DMF were located in the southeastern direction, that is, plants A, B, and C. Further investigations were still required to clarify which facilities were the dominant odorant sources, and these are described below.

Application of Mobile OPFTIR Systems
To identify the potential odorant sources and narrow down the facilities that need to be further investigated, a pair of mobile OPFTIR systems was first deployed in the upwind and downwind locations of plant A (see Fig. 3) for 70 hours.The pollution rose plot shows that the monitoring path of the mobile OPFTIR system at the upwind location found higher DMF concentrations and detection frequencies when the prevailing wind directions were south, revealing that more significant DMF sources were located upwind of plant A. In other words, the amount of DMF emitted from the facilities at plant A was insignificant while compared to the amount emitted upwind of its location.
The pair of mobile OPFTIR systems was then deployed in the upwind and downwind locations of plant B for 120 hours.The pollution rose plots derived from the southern mobile OPFTIR system showed that there were significant DMF sources located in the northwestern direction.In addition, the pollution rose plot derived from the northwest mobile OPFTIR system showed that its monitoring path observed high DMF concentrations when the wind was blown from the southeast.In other words, at least one significant DMF source was located in the territory of plant B.
Then the pair of mobile OPFTIR systems was deployed in the upwind and downwind locations of plant C for 100 hours.The pollution rose plot derived from the upwind mobile OPFTIR system showed there were DMF sources located in the northwest direction.By comparing the DMF concentrations and detection frequencies obtained from the pair of mobile OPFTIR systems, no significant increase in atmospheric DMF was observed across the territory of plant C, and thus the amount of DMF emitted from the facilities of plant C was insignificant compared to the amount emitted upwind of its location.Finally, the pollution rose plots derived from the pair of mobile OPFTIR systems deployed in downwind location of plant D for 70 hours verify the existence of DMF sources located in the territory of plant B, and showed their influences on the atmospheric DMF concentrations, which decreased with increasing distance from plant B.
Based on the monitoring results obtained from the mobile OPFTIR systems, we concluded that there was at least one significant DMF source located in the territory of plant B. Therefore, in the next step, the same approach of using a pair of mobile OPFTIR systems was applied to identify the DMF sources, but this time with a smaller area and more specific facilities.

Identification of DMF Sources
There were several potential DMF sources in plant B, including evaporative emissions from manufacturing processes which used DMF as a raw material, stack emissions from a regenerative thermal oxidizer (RTO) and fugitive emissions from pipelines and tanks for recovery of DMF and toluene.The detailed raw materials and products of the manufacturing processes related to DMF (M1, M2, M3, M4 and M5) are listed in Table S4, including the dry and wet polyurethane (PU) synthetic leather processes, PU resin and synthetic fiber manufacturing processes.
We thus divided Plant B into three sections for more detailed investigation, as shown in Fig. S1, based on the spatial distributions of the three potential DMF sources.Similarly, the pair of mobile OPFTIR systems was simultaneously deployed at the upwind and downwind locations of each of the three sections at Plant B. The first and second monitoring paths were deployed to monitor the difference in odorant concentrations at the upwind and downwind locations in the first section, the second and third monitoring paths were used for those in the second section, and the second and fourth monitoring paths were applied for those in the third section.The prevailing wind directions were northwest during the first section investigation, northwest and southeast during the second section investigation, and southeast and southwest during the third section investigation, respectively.The pollution rose plots of DMF derived from the pair of mobile OPFTIR systems deployed in the first section of plant B are shown in Fig. 4. The monitoring path of the mobile OPFTIR system situated at the downwind location (the second monitoring path) observed significant DMF concentrations and detection frequencies in the northwest direction, while that situated at the upwind location (i.e., the first monitoring path) detected DMF only when the wind blew from the southeast.
The data gathered during the northwestern or southeastern wind periods were averaged to further show the differences in DMF concentrations among the monitoring paths of the mobile OPFTIR systems.When the prevailing winds were blown from the northwest, the average atmospheric DMF concentration detected in the downwind monitoring path was 1600 ppb, which was approximately 800 times higher than that in the upwind monitoring path.However, the differences in DMF concentrations were insignificant between the upwind and downwind monitoring paths (521 ppb vs. 447 ppb) when the prevailing winds were blown from the southeast.These results revealed that the evaporative emissions from the M1 production process were a significant DMF emission source.The evaporative emission source should be close to the second monitoring path because the wind speed of the southeastern wind was close to calm, and the slightly higher DMF concentration observed in the upwind monitoring path (521 ppb) may be due to the effect of the dispersion.
The pollution rose plots of DMF derived from the pair of mobile OPFTIR systems in the second section of plant B are shown in Fig. 5.When the prevailing winds were blown from the northwest, the monitoring path of the mobile OPFTIR system situated at the upwind location (i.e., the second monitoring path) observed significant DMF concentrations and detection frequencies, while that situated at the downwind location (i.e., the third monitoring path) detected relatively lower DMF concentrations and detection frequencies.The

First monitoring path
Second monitoring path average atmospheric DMF concentration detected in the upwind monitoring path was 1360 ppb, 5.6 times higher than that in the downwind monitoring path.When the prevailing winds turned to the southeast, the monitoring path of the mobile OPFTIR system situated at the upwind location (i.e., the third monitoring path) still observed higher DMF concentrations and detection frequencies than that situated at the downwind location (i.e., the second monitoring path).
The average atmospheric DMF concentration detected in the upwind monitoring path was 1530 ppb, three times higher than that in the downwind monitoring path.These results revealed that the potential DMF sources in the second section (including RTO and DMF recycling plants) were not significant compared to the evaporative emissions from the M1 process.Furthermore, there must also be other potential DMF source(s) in the southeast, outside the second section.
The pollution rose plots of DMF derived from the pair of mobile OPFTIR systems in the third section of plant B are shown in Fig. 6.When the prevailing winds were blown from the southeast and southwest, the monitoring path of the mobile OPFTIR system situated at the downwind location (i.e., the second monitoring path) observed much higher DMF concentrations and detection frequencies than that situated at the upwind location (i.e., the fourth monitoring path).The average atmospheric DMF concentration detected in the downwind monitoring path was 430 ppb, 29 times higher than that in the upwind monitoring path.When the prevailing winds were blown from the northwest, the monitoring path of the mobile OPFTIR system situated at the upwind location (i.e., the second monitoring path) observed higher DMF concentrations and detection frequencies than that situated at the downwind location (i.e., the fourth monitoring path).However, the differences in the DMF concentrations were not significant between the upwind and downwind monitoring paths (528 ppb vs. 477 ppb).
These results from the sequential deployments of the pair of mobile OPFTIR systems in different sections revealed that there was no significant DMF source outside the territory of plant B. Furthermore, in addition to the evaporative emissions from the M1 production process, there were other DMF sources.These DMF sources could be the evaporative emissions from a cluster of the production

Relationship between DMF and other VOCs
Besides DMF, many other VOCs were detected by the mobile OPFTIR systems, such as methane, cyclohexane, ammonia, 1,3-butadiene, 2-butanone, ethyl acetate, ethylene, methanol, propylene, styrene, toluene, vinyl acetate, propane, butane, chlorodifluoromethane, isopropanol, methyl acetate, acrylonitrile, hydrogen cyanide, 2-methyl butane, nitrogen trifluoride, trichlorofluoromethane, methyl acetate, nitrogen trifluoride, trichlorofluoromethane and cyclohexanone.However, only four odorants, 2-butanone, ethyl acetate, toluene, and isopropanol, had detection frequencies greater than 1%.Tables S1-S3 in the Supporting Information list the percentages of these concentrations exceeded the related air quality regulation and odor thresholds for DMF and these four odorants for the three sections of plant B. The percentages of 2-butanone, ethyl acetate, and DMF concentrations over the related odor thresholds were all more than 25%.
To clarify if the odorant sources of DMF (i.e., M1, M2, M3, M4 and M5) were also responsible for the results with regard to 2-butanone, ethyl acetate, toluene, and isopropanol, Spearman correlation analyses were conducted to evaluate the relationships between DMF and other four odorants.Only data obtained from the mobile OPFTIR situated at the downwind location of the DMF emission sources are included in the correlation analyses, that is, the data from the first and third monitoring paths when the prevailing winds were blown between east-southeast and south-southeast, as well as the data from the second and fourth monitoring paths when the prevailing winds were blown between westnorthwest and north-northwest.The results of the correlation analyses are summarized in Table 1.From the monitoring paths situated at the downwind location of the M1 process, we found that 2-butanone, ethyl acetate, and isopropanol were significantly correlated with DMF, and their correlation coefficients were in the range of 0.322-0.731,0.493-0.794and 0.112-0.403,respectively.Therefore, besides DMF, M1 was also the odorant source for 2-butanone, ethyl acetate, and isopropanol.However, with regard to the monitoring paths situated at the downwind location of the cluster of M2, M3, M4 and M5 processes, none of the four odorants was found to be significantly correlated with DMF.Unlike M3, M4 and M5, DMF was the only solvent used in the M2 process.We thus speculated that M2 was responsible for the high atmospheric DMF concentrations observed in the third monitoring path.
In some circumstances, when the prevailing winds were blown between east-southeast and south-southeast, the concentrations of 2-butanone, ethyl acetate, and isopropanol from the upwind location of the M1 process, that is, data from the second monitoring path, were still significantly correlated with the DMF concentrations (with correlation coefficients ranging from 0.297 to 0.931).These results revealed that the evaporative emissions from the M1 process should be close to the second monitoring path.

CONCLUSIONS
The aim of this study was to develop an integrated approach to monitor and identify the sources of odorants emitted from a petrochemical industrial complex by applying the stationary and multiple mobile OPFTIR systems.The mean atmospheric DMF concentration in the monitoring path of the mobile OPFTIR system situated downwind of the M1 process (dry PU synthetic leather process) was 1600 ppb, 800 times higher than that in the upwind monitoring path.Similarly, the mean atmospheric DMF concentration in the monitoring path situated downwind of the cluster of M2, M3 and M5 was 430 ppb, 29 times higher than that in the upwind monitoring path.The results revealed that the evaporative emissions from these production processes were a significant source of DMF emissions.In addition to DMF, there were many other VOC compounds that were detected by the mobile OPFTIR systems.By conducting correlation analyses on the data obtained from the mobile OPFTIR situated at the downwind location of the DMF emission sources, the dry PU synthetic leather process was found to be the odorant source of 2-butanone, ethyl acetate, and isopropanol, as well as DMF, while DMF was the sole odorant emitted from M2 process (wet PU synthetic leather process).

Fig. 1 .
Fig. 1.The investigated petrochemical complex and the location of the stationary OPFTIR system.The blocks marked A, B and C are factories involved in DMF.

Fig. 2 .
Fig. 2. The temporal trends of DMF detected by the stationary OPFTIR system in 2012.

Fig. 3 .
Fig. 3. Wind rose and pollution rose plots of DMF in the investigated petrochemical industrial complex derived from the pair of mobile OPFTIR systems.

Fig. 4 .Fig. 5 .
Fig. 4. Pollution rose plots of DMF derived from the pair of mobile OPFTIR systems in the first section of plant B.

Fig. 6 .
Fig. 6.Pollution rose plots of DMF derived from the pair of mobile OPFTIR systems in the third section of plant B.

Table S1
Detection limit of the mobile OPFTIR TableS2The coefficient of variation and accuracy of the mobile OPFTIR for five