{"responseHeader":{"status":0,"QTime":10,"params":{"q":"{!q.op=AND}id:\"14408\"","hl":"true","hl.simple.post":"","hl.fragsize":"5000","fq":"!embargo_tdt:[NOW TO *]","hl.fl":"ocr_t","hl.method":"unified","wt":"json","hl.simple.pre":""}},"response":{"numFound":1,"start":0,"docs":[{"type_t":"Event","spatial_coverage_t":"Houston, Texas","contributor_t":"Ponchaut, N.; Kytomaa, H.","ark_t":"ark:/87278/s6ps0t4c","thumb_s":"/f5/52/f55261eed5a11e93f06a6c54e6a9a74766d1a428.jpg","oldid_t":"AFRC 14490","setname_s":"uu_afrc","subject_t":"2014 AFRC Industrial Combustion Symposium","restricted_i":0,"rights_t":"No copyright issues exist.","format_t":"application/pdf","creator_t":"Marr, K.","date_t":"2014-09-08","modified_tdt":"2015-10-29T00:00:00Z","conference_t":"AFRC 2014 Industrial Combustion Symposium","description_t":"Paper from the AFRC 2014 conference titled Emissions Estimation Methodologies for Industrial Flares by K. Marr.","file_s":"/b3/90/b39082dbe2994101ad4c772216d4942fb240cc19.pdf","title_t":"Emissions Estimation Methodologies for Industrial Flares","abstract_t":"Flares are essential in safely managing excess combustible gases that can result from normal operations as well as process upsets and emergency conditions in refineries, chemical plants, and other industrial facilities. Government regulations require that facilities inventory the emissions from flaring events. However, direct measurement of flare emissions is difficult and often impractical. Current practices for estimating flare emissions often incorporate a combination of several methodologies. Common methodologies include applying emissions factors as specified by AP-42 or assuming a 98% combustion efficiency if the flare is operating in accordance to 40 CFR 60.18. In general, these methods have been developed for properly operating flares that are smokeless, not over-steamed or over-aerated. For scenarios where a flare may not be properly operating, these methods cannot be used. This work overviews methodologies including statistical and flare scaling analyses that can be used to estimate emissions from flares operating both in and outside of the operation criteria in 40 CFR 60.18. Such scenarios include under or over-steaming and high turndown conditions.","id":14408,"created_tdt":"2015-10-28T00:00:00Z","parent_i":0,"_version_":1642982434986262528,"ocr_t":"EMISSIONS ESTIMATION METHODOLOGIES FOR INDUSTRIAL FLARES Kevin C. Marr, Ph.D., P.E. kmarr@exponent.com Nicolas F. Ponchaut, Ph.D., P.E. nponchaut@exponent.com Francesco Colella, Ph.D. fcolella@exponent.com Harri K. Kytömaa, Ph.D., P.E. hkytomaa@exponent.com Exponent, Inc., 9 Strathmore Road, Natick, MA 01760, USA ABSTRACT Flares are essential in safely managing excess combustible gases that can result from normal operations as well as process upsets and emergency conditions in refineries, chemical plants, and other industrial facilities. Government regulations require that facilities inventory the emissions from flaring events. However, direct measurement of flare emissions is difficult and often impractical. Current practices for estimating flare emissions often incorporate a combination of several methodologies. Common methodologies include applying emissions factors as specified by AP-42 or assuming a 98% combustion efficiency if the flare is operating in accordance to 40 CFR 60.18. In general, these methods have been developed for properly operating flares that are smokeless, not over-steamed or over-aerated. For scenarios where a flare may not be properly operating, these methods cannot be used. This work overviews methodologies including statistical and flare scaling analyses that can be used to estimate emissions from flares operating both within and outside of the operation criteria specified by 40 CFR 60.18. Such scenarios include under or over-steaming and high turndown conditions.INTRODUCTION Flares are common safety devices that are used to manage waste gases that are produced from normal or emergency operations in refineries and other industrial facilities. EPA regulations, 40 CFR 60.18, require that flares operate without visible emissions, i.e. smoke. One common method to suppress smoke in modern flare designs is to add steam to the vent gas upstream of the reaction zone. In the 1980s, several benchmark studies were performed that formed the basis for current EPA regulations [1] [2], [3], [4]. These studies identified several parameters that affect combustion efficiency. They found that when a flare operated near its flame stability limits, the combustion efficiency degraded significantly. More recent studies have shown that several additional parameters can have an influence on combustion efficiency, including hydrogen content, excess steam addition and crosswind [5], [6]. These have also shown that even if a flare is operating within the envelope specified by EPA regulations, excessive steam addition can result in destruction efficiencies that fall below the assumed 98% threshold [6], [7]. Flares operating within the EPA specified envelope can also have destruction efficiencies higher than 98%. Although EPA threshold parameters provide simple guidelines in part to maintain the flare destruction efficiency above 98%, they do not provide, nor were they intended to provide, quantitative predictions of flare emissions. In addition to over-steaming scenarios, a flare may operate outside the EPA regulations as a result of either a failure of the air or steam assist system or operator error. In these scenarios, where the flare is over-steamed or over-aerated, or when the flare is operating outside the EPA regulated operating envelope, there is little guidance on estimating the flare emissions. In addition to regulatory emissions reporting, estimation of the emissions source from a flare is the first step in predicting or investigating ground level impact of a flaring event. Unintended or emergency flaring events are often accompanied with a need to assess the impact of the pollutants on the surrounding environment. Several modeling tools such as 3-dimensional Computational Fluid Dynamics (CFD) or integral plume dispersion models used to analyze these events can provide quantification of ground level impacts, but they require a quantitative estimate of the pollutant emissions as an input to the dispersion model. Common estimation methods include applying emissions factors specified by AP-42 or assuming a 98% destruction efficiency for flares operating in accordance to 40 CFR 60.18. These methods, however, are limited for scenarios where the flare is operating within the design envelope. The scope of this work is to present several methods that can be used to predict emissions from an arbitrary flaring event. A review of current standard methodologies and a discussion of additional methods will be presented. FLARE EMISSIONS ESTIMATION PRACTICES Flare emissions estimation methodologies can be organized in a hierarchy based on the accuracy of the method [8]. The hierarchal ranking of the methods is based on the fidelity of the estimation, where the number of uncertain parameters decreases for higher ranked methods. Table 1 summarizes the input parameters for the estimation methodologies. Table 1: Flare Emissions Estimation Methodology Inputs [8] Rank Method Description Directly Measured Parameters Estimated Parameters 1 • Direct measurement of emissions from the plume • Plume species concentrations • Species composition of the flare gas • Flow rate • Plume entrainment • 2 • Continuous composition monitoring of flare gas • Continuous flow rate monitoring • Species composition of flare gas • Flow rate • Combustion efficiency 3 • Periodic composition monitoring of flare gas • Continuous flow rate monitoring • Species composition samples from flare gas • Flow rate • Sample is representative of species composition • Combustion efficiency 4 • Continuous flow rate monitoring of flare gas • Continuous heating value monitoring • Flow rate • Heating Value • Combustion efficiency • Emissions factors 5 • Engineering calculations • Details of process units that feed the flare • Combustion efficiency • Process flow ratesFor each methodology class, a set of inputs parameters are required. The parameters are either directly measured or estimated using engineering analysis. For example, flow rate of fuel supplied to the flare can be a measured quantity if the flare is equipped with flow monitors. However, if flow monitors are unavailable or non-functioning, the flow rate can be estimated based on engineering analysis of the upstream process conditions. Recent developments in remote gas sensing and imaging has led to techniques where the emissions from a flare can be directly measured. Several techniques including Differential Absorption Lidar (DIAL) and Passive and Active Fourier Transform Infrared spectroscopy (PFTIR, AFTIR) are spectroscopic methods that are non-intrusive and do not require a probe to be positioned downstream of the flare. These techniques, categorized as Rank 1 in Table 1, can be used to determine the concentration of gas species in the flare plume or to directly measure the combustion efficiency of the flare. However, for emissions inventories, the mass emission of a particular pollutant is required, rather than the concentration of that pollutant in the plume. To determine the mass emission of a particular pollutant species from a direct measurement of the pollutant concentrations, the air entrainment needs to be quantified either by direct measurement, if possible, or engineering analysis. A second method to determine the mass emission is to calculate the combustion efficiency from the combustion product concentrations in the plume. If the gas composition of the flare is known, then the species mass emission can be estimated directly from the measured combustion efficiency and flow rate. Several testing programs have used this method of measuring the combustion efficiency to determine flare emissions. The results of these testing campaigns are summarized in a recent EPA issued report, Parameters for Properly Designed and Operated Flares [9]. However, these direct measurement techniques have not been implemented in the majority of petroleum and chemical refineries. Rank 2 and 3 methodologies consist of monitoring the flare gas composition and flow rate of the fuel gas supplied to the flare. For Rank 2 methodologies, the gas composition is continuously monitored, typically with an online gas chromatography-mass spectrometry system. These systems can sample the gas at intervals on the order of several minutes. For Rank 3 methodologies, the gas is sampled periodically, either daily or weekly, and analyzed offsite. For both ranks, the flare gas concentration is measured upstream of the flare, and the combustion properties of the flare must be assumed. The destruction efficiency of the flare is estimated based on either previous testing of the flare or engineering analysis. Rank 2 and 3 methodologies are primarily applied toward VOC, SO2, or other species that are assumed to be consumed, or destroyed by the flare. In Rank 4 methodologies, the flow rate and heating value of the flare gas are monitored, but the flare gas species are not monitored Predetermined emission factors are then applied depending on the flow rate and heating value using the following equation [8]: ̇= × × [Eqn 1] The emissions mass rate of a pollutant species, ̇, for a specific period is determined from the volume of gas sent to the flare during each measurement period, , the net heating value of the flare gas stream, , and the emissions factor of the pollutant species, . The most widely used emission factors for hydrocarbons, CO, NOx and soot are specified in AP-42 Section 13.5.1 [10]. For scenarios where the required parameters specified in Rank 1-4 methodologies are unknown, a variety of engineering analyses, loosely categorized as Rank 5 methods, can be applied. The required parameters are often determined from knowledge of the process units that supply gas to the flare. Temperature, pressure, gas composition and other properties of the process units can be useful parameters in determining the properties of the flare gas. For example, the position of valves that meter process streams to the flare and the pressure of the upstream process can be used to estimate the flow rate using standard orifice flow equations. Another such method proposed by RTI International estimates the emissions rate by applying emissions factors based on the overall capacity of a refinery [11]. EMPIRICAL CORRELATIONS FOR COMBUSTION EFFICIENCY Many of the methods discussed above estimate emissions based on the determination of the flare combustion efficiency. However, in most refineries, direct measurement of combustion efficiency is not possible and must be estimated. In the past few years, EPA enforcement of the Clean Air Act has led to the testing of flares at several industrial facilities. In 2012, the EPA issued a report, Parameters for Properly Designed and Operated Flares, which analyzes results from nine elevated flare tests [9]. Table 2 summarizes the different studies analyzed by the EPA report. The measurement techniques used by these studies include extractive measurements from probes (Extractive), and non-intrusive methods (PFTIR, AFTIR). Table 2: Test reports included in EPA analysis Study ID Authors Date % H2 in Vent Gas Test Method EPA-600/2-83-052 McDaniel [1] July 1983 0 Extractive EPA-600/2-85-106 Pohl and Soelberg [3] Sept 1985 0 Extractive MPC TX Clean Air Engineering [12] May 2010 3.1-24 PFTIR INEOS Clean Air Engineering [13] July 2010 0 PFTIR MPC Detroit Clean Air Engineering [14] Nov 2010 7.0-55 PFTIR FHR (AU) Clean Air Engineering [15] June 2011 13-47 PFTIR FHR (LOU) Clean Air Engineering [15] June 2011 20-30 PFTIR SDP EPF Shell Global Solutions [16] Apr 2011 37-62 PFTIR TCEQ Allen and Torres [7] Aug 2011 0 Extractive, AFTIR, PFTIR Using the results from these tests, empirical correlations can be used to estimate the combustion efficiency of an arbitrary flaring event. The combustion efficiency was shown to have a strong correlation with the net heating value of the combustion zone gases, (NHVCZ). For a steam-assisted flare, where steam is also present in the combustion zone, a net heating value of the combustion zone gases (NHVCZ) in BTU/scf can be defined as follows [7]: =̇/+̇/̇/+̇+̇/ [Eqn 2] where the net heating value, NHV, mass flow rate, ̇, and molecular weight, MW, for the vent gas, pilot gas, and steam are indicated by the subscripts VG, PG and S, respectively. Note that Equation 2 requires consistent units. For English units, the net heat value is in BTU/scf, the mass flow rate is in lb/hr and the molecular weight is in lb/lb-mole. Figure 1 shows a plot of the measured combustion efficiency for the flares considered in the testing studies detailed in Table 2, and clearly shows a sharp decrease in combustion efficiency when NHVCZ < 300 BTU/scf. Figure 1: Measured combustion efficiency for industrial-scale flares Previous work presented two such methods-a deterministic method and a statistical method based on a Monte Carlo scheme [17]. A brief summary will be included here. Generalized Scaling Previous work showed that the correlation between NHVCZ and combustion efficiency can be improved by apply a hydrogen correction [6], [15], [17]. A general form for hydrogen correction can be defined as follows: =100−1 [Eqn 3] =+2 [Eqn 4]where X and Y are the scaled variables for NHVCZ and the combustion inefficiency, 100-CE, respectively. The X and Y scaled variables can be interpreted as the effective NHVCZ and the effective combustion inefficiency, respectively. For the hydrogen correction, the scaling functions are assumed to be functions of the hydrogen molar ratio in the flare combustion zone, RH2. The functions f1 and f2 are assumed to be of the form 1=(1−2)(1− 2) [Eqn 5] 2= (2) [Eqn 6] where D, E, and F are constant coefficients. Both f1 and f2 are chosen such that when the vent gas does not contain hydrogen, the scaled variables, X and Y, return to their dimensional forms, NHVCZ and 100-CE. Similarly, f1 was also chosen so that when pure hydrogen is flared, the combustion efficiency would tend to 100%. Table 3 summarizes the expressions for f1 and f2 for the hydrogen adjusted NHVCZ correction and the optimized generalized scaling proposed in Ref. [17]. Figure 2 shows the effect of the hydrogen scaling correction. The black symbols represent flares where the fuel did not contain hydrogen. For the flares that did contain hydrogen, the unadjusted NHVCZ is show in blue and the hydrogen-adjusted NHVCZ is show in red. The arrows show the shift for a few example data points that result from the hydrogen scaling correction. Table 3: Correction functions f1 and f2 for the hydrogen adjusted NHVCZ and the generalized scaling corrections [17] Correction f1 f2 No adjustment 1 0 Generalized Scaling (1-RH2)(1+4.2RH2) 169.520.7Figure 2: Flare test data plotted with respect to the hydrogen adjusted NHVCZ correction. Examples of the resultant shift due to the correction are indicated by purple arrows. [17] Deterministic Method The deterministic approach to predict flare emissions is to find a best fit curve for the test data. Assuming an exponential function of the form Ae(-Bx)+C, a best fit curve can be determined for the generalized scaling (Figure 3). The simplicity of a simple exponential curve fit is appealing as a practical tool. However, other curve fit methodologies can provide a better statistical fit. One such method is to apply a logarithmic transform to the y-axis, which in this case is the combustion efficiency. Another method is to apply a weight-averaged curve fit, where certain data points are give more weight, or influence, on the curve fit. In general, the specific weight assigned to the individual data points is based on some assumed distribution function.Figure 3: Best fit curve for the generalized scaling correction for hydrogen [17] Regardless of the details of the curve fit, deterministic prediction methods have limitations. One of the main limitations is that any curve fit assumes an arbitrarily chosen shape of the fitting function. In addition, the way the fit is obtained (i.e., using weights or logarithm transforms, etc.) also influences the results. Deterministic approaches based on empirical data are only able to provide an emissions estimate of an \"average\" flare. To provide insight for cases where the flare may not behave like an \"average\" flare, determination of upper and lower bounds of the emissions estimate is necessary. Uncertainty analysis for deterministic methods is limited to bounds on the fit coefficients and provides no information on the predictive uncertainty of the aggregate emissions from a flaring event. Monte Carlo Prediction Method The Monte Carlo prediction method is a statistical method that provides an estimate of the emissions and confidence bounds. The Monte Carlo algorithm is divided in two portions.• First, for a flaring event, the effective NHVCZ values are calculated at a particular time interval (hourly, daily, etc.). At each time, a net heating value bin is defined. This bin is centered on the calculated NHVCZ at the time of interest, plus and minus a given net heating value interval (bin half width). All the combustion efficiency data points published in EPA reports that fall into the bin are then considered as possible representative emissions characteristics at the time of interest. The method assumes that all the combustion efficiencies within that bin are possible and equally probable. Figure 4 illustrates a bin that is centered at 350 BTU/scf. • Once the list of possible combustion efficiency values is determined, the Monte Carlo method selects randomly one data point in each bin. In the example shown in Figure 4, each data point in the bin window has an equal probability of being selected if the effective NHVCZ is 350 BTU/scf. Based on this random sampling, the flare emissions are calculated. A single trial would not be meaningful since it would only be one of the possible outcomes, and therefore the process is repeated many times, which allows the method to determine an expected average outcome, as well as probability ranges around the average.Figure 4: Example of a 25 BTU/scf bin centered at an effective NHVCZ of 350 BTU/scf. Inset plot shows a zoomed in view of the bin. Other than the choice of bin size, the Monte Carlo method does not make any additional assumptions such as the form of a fit function, and therefore eliminates the bias that a deterministic curve fitting method introduces. In addition, the data scatter associated with the test data is implicitly included in the Monte Carlo scheme. As a result of the random selection of the data points, the Monte Carlo method also directly determines the statistical distribution for the emissions prediction from the multiple random trials for each time period of interest. From the statistical distribution of the random trials, confidence intervals can be defined. The ability to determine confidence intervals is the main advantage of the Monte Carlo method over the deterministic method. Not only is the most probable emissions from a flare predicted, but also the upper and lower bounds defined by the confidence intervals. Figure 5 shows an example of the validation results of the Monte Carlo method presented in Ref. [17]. In the validation study, emissions from a testing site was predicted using the Monte Carlo and deterministic methods and compared to the emissions calculated from the measured combustion efficiency . For the example shown in Figure 5, the FHR AU test site was chosen. A flaring event was created from the flow rate data recorded during the FHR AU testing by assuming that each flow rate condition corresponded to one hour of a flaring event. The emissions from the FHR AU flaring event were predicted by applying the prediction methods to the measured combustion efficiency data from the remaining eight test sites. The results from both the Monte Carlo (shown in blue) and the deterministic (shown in red) methods were compared to the actual emissions from the FHR AU test site. The varying shades of blue show the probability distribution of the predicted emissions. In Figure 5, the Monte Carlo method predicts that there is a 99% chance that the actual cumulative emissions are below 11 tons. Similarly, 1%, 5%, 10%, 90%, and 95% probabilities are also shown. An alternative interpretation is that the dark blue area between the 10% and 90% contours define the 80% prediction interval where there is an 80% probability that the actual emissions fall within the interval. Similarly, the areas between the 5% and 95% probabilities and the 1% and 99% probabilities define the 90% and 98% prediction intervals, respectively. The deterministic method considered is a simple best fit exponential function.Figure 5: Comparison of the Monte Carlo and the deterministic predictions of C5+ emissions for FHR AU test data using generalized scaling correction for hydrogen. The prediction intervals are shown for the Monte Carlo analysis. [17]. In the Monte Carlo method, the accuracy of the prediction increases as the number of predicted events increase. In practice, the Monte Carlo method would be beneficial in estimating cumulative emissions from longer events that may last over several days. However, care should be taken when applying the Monte Carlo method to flaring events. The accuracy of the Monte Carlo method is also dependent on the scatter in the combustion efficiency measurements from the nine test sites. In the validation case presented in Ref [17], the scaling of the data with NHVCZ does not consider other parameters that may affect the combustion efficiency. This results in an inherent bias which, in this case, leads to an under-prediction of the cumulative emissions. A determination of the appropriateness of the combustion efficiency test data must be made prior to applying the Monte Carlo method to a subject flare. The accuracy of the prediction is directly related to whether a subject flare can be described by the probability distribution of the existing combustion efficiency test data. However, the strength of the Monte Carlo method is that it provides confidence intervals. Even though the FHR AU validation case under predicts the cumulative emissions, the actual emissions do fall within a reasonable confidence interval. Using even a 95% confidence interval can provide reasonable upper and lower bounds for emissions. To improve the prediction accuracy of the Monte Carlo method, the scaling of the measured combustion efficiencies needs to be further refined. The current proposed scaling does not account for variable operating conditions (i.e. variable flow rates and composition, over-steaming and over-aerating), and atmospheric conditions (i.e. variable wind speed and direction, and atmospheric stability). These parameters may have an effect on the scatter in the measured combustion efficiency data, and a proper determination of these effects may require significantly more data. Developing such a database experimentally may not be practical due to the expense and difficulties associated with large-scale testing. Such parametric studies may be done more efficiently with CFD modelling. Due to the complexity in flare fuel mixtures and large range of fluid flow scales associated with industrial flares, direct prediction of emissions species from a flare is not possible with the current state of available commercial CFD packages. However, CFD can provide insight in understanding and determining the sensitivity of each flare parameter. For example, simulations of the non-reacting flow field of a flare can be used in to investigate the effect of steam injection. Many steam-assisted flares are designed to inject steam at two locations through steam nozzles in the flare stack (center steam) or above the flare stack prior to the combustion zone (upper or ring steam). In the analysis presented here, the individual steam flow rates from the center and upper injection locations were summed, and the resultant total steam flow rate was considered in the scaling analysis. However, the ratio of steam delivered to the center or upper nozzles has an effect on the combustion efficiency. By understanding the relative effectiveness of the upper and lower steam injection in mixing the flare fuel and steam, it is possible that a more refined scaling could be developed. CONCLUSIONS Several methods can be applied to estimate emissions from flares. These methods can be ranked according to the uncertainty in the required input parameters. However, nearly all methods of quantifying emissions from flares require an estimate of the combustion characteristics of the flare. Current methods used in practice often apply emissions factors based on an assumption that the flare is operating at high efficiency as long as the flare operates in accordance to 40 CFR 60.18. However, for scenarios where the flare is not operating efficiently, such as over-steamed or over-aerated conditions, or during incidents where the flare is operating at efficiency above the assumed 98% threshold, emission factor based estimates may not be accurate. Empirical correlations can be useful analysis tools to determine the combustion efficiency of a flare. Although deterministic methods are easy to implement, statistical methods can provide more accurate estimates, and more importantly, uncertainty bounds. The ability of statistical methods, such as the Monte Carlo method reviewed here, to predict not only the most probable emission, but also provide upper and lower bounds for the predicted emission is the primary advantage of statistical methods over simple deterministic methods. However, the accuracy of the estimate will be dependent on how similar the flare of interest is to the flares used to develop the empirical correlations. Also, the accuracy of the statistical method will be higher for larger empirical datasets. In addition to regulatory reporting, emissions estimates are important in assessing impact of the flare plume on the surrounding environment and people. Models such as plume dispersion models or CFD are often used to simulate flaring events and determine the ground level pollutant concentrations in neighboring areas. However, these models require an a priori knowledge of the species emitted from the flare. The Monte Carlo method summarized in this work provides a frame work to not only provide a source term for the model but also upper and lower bounds for worst case and best case scenarios. REFERENCES [1] M. McDaniel, \"Flare efficiency study,\" EPA-600/2-83-052, 1983. [2] J. Pohl, R. Payne and J. Lee, \"Evaluation of the efficiency of industrial flares,\" EPA-600/2-84-095, 1984. [3] J. Pohl and N. Soelberg, \"Evaluation of the efficiency of industrial flares: flare head design and gas composition,\" EPA-600/2-85-106, 1985. [4] J. Pohl and N. Soelberg, \"Evaluation of the efficiency of industrial flares: H2S gas mixtures and pilot assisted flares,\" EPA-600/2-86-080, 1986. [5] P. Gogolek and A. Caverly, \"Emissions from elevated flares-a survey of the literature,\" CanmetENERGY, 2009. [6] S. Evans, \"Insights from passive FTIR flare performance testing,\" in AFRC Industrial Flares Colloquium, Houston, TX, 2011. [7] D. Allen and T. V.M., \"TCEQ 2010 Flare Study Final Report,\" PGA No. 582-8-86245-fy-09-04, The Unversity of Texas at Austin, 2011. [8] RTI International, \"Emission Estimation Protocol for Petroleum Refineries,\" Prepared for US Environmental Protection Agency, Office of Air Quality Planning and Standards, Research Triangle Park, NC, 2011. [9] US EPA Office of Air Quality Planning and Standards, \"Parameters for properly designed and operated flares,\" US EPA, 2012. [10] Office of Air Quality Planning and Standards, \"Compilation of Air Pollutant Emission Factors, Volume 1: Stationary Point and Area Sources,\" US Environmental Protection Agency, Research Triangle Park, NC, 1995. [11] RTI International, \"Petroleum Refinery Source Characterization and Emission,\" Prepared for US Environmental Protection Agency, Office of Air Quality Planning and Standards, Research Triangle Park, NC, 2002. [12] Clean Air Engineering, Inc., \"Performance test of a steam-assisted elevated flare with passive FTIR,\" Marathon Petroleum Company, LLC, 2010. [13] INEOS ABS (USA) Corporation, \"Passive Fourier transform infraed technology (FTIR) evaluation of P001 process control device at the INEOS ABS (USA Corporation),\" INEOS ABS, Addyston, OH, 2010. [14] I. Clean Air Engineering, \"Performance test of a steam-assisted elevated flare with passive FTIR-Detroit,\" Marathon Petroleum Company, LLC, 2010. [15] Clean Air Engineering, Inc., \"PFTIR test of steam-assisted elevated flares-Port Arthur,\" Flint Hills Resources Port Arthur, LLC, 2011. [16] Shell Global Solutions (US) Inc., \"Shell Deer Park Refining LP Deer Park Refinery East Property flare test report,\" Shell Gobal Solution (US) Inc., Houston, TX, 2011. [17] K. Marr, N. Ponchaut and H. Kytomaa, \"Analysis of Combustion Efficiencies for Inustrial Steam-Assisted Flares,\" in AFRC 2013 Industrial Combustion Symposium, Koloa, HI, 2013. [18] C. Law, Combustion Physics, New York: Cambridge University Press, 2006. [19] C. Baukal, The John Zink Combustion Handbook, New York: CRC Press, 2001. [20] \"Flare Peer Review Panel Comments,\" 2012. [Online]. Available: http://www.epa.gov/airtoxics/flare/2012flarepeerreviewmemo.pdf."}]},"highlighting":{"14408":{"ocr_t":[]}}}