Comparison of Remote Sensing and Extractive Sampling

May 16, 2012 - Several remote sensing technology developers also participated in the study and a comparison of their performance to extractive samplin...
0 downloads 5 Views 2MB Size
Article pubs.acs.org/IECR

Comparison of Remote Sensing and Extractive Sampling Measurements of Flare Combustion Efficiency Joda Wormhoudt,*,† Scott C. Herndon,† Jon Franklin,† Ezra C. Wood,†,§ Berk Knighton,‡ Scott Evans,⊥ Curtis Laush,∥ Mark Sloss,∥ and Robert Spellicy∥ †

Aerodyne Research, Inc., Billerica, Massachusetts Department of Chemistry and Biochemistry, Montana State University, Bozeman, Montana ⊥ Clean Air Engineering, Palatine, Illinois ∥ Industrial Monitor & Control Corp., Round Rock, Texas ‡

ABSTRACT: The 2010 Comprehensive Flare Study provided the opportunity for the first blind validation of a remote sensing technique for flare combustion efficiency (CE) against extractive analysis techniques. The overall test results show that both remote sensing and extractive sampling accurately determined the flare performance curve. Both remote and extractive sampling techniques are challenged by the fluctuating and inhomogeneous nature of the flare exhaust plume. Accurate measurement of CE values near 100% is of interest to flare manufacturers, users, and regulators, while measurement of low CE flares is of interest in the development of other applications of combustion monitoring, and the 2010 tests provided information on both. In practice, accurate values of CE can be determined through the measurement of a small number of gaseous species, including fuel components and products of combustion. Nominal error bars generated from the fluctuations in these component measurements were adequate to account for most of the differences between the remote and extractive sampling CE measurements. The additional analysis reported here focused on individual species measurements and on cases where CE values measured by the two techniques differed by more than the nominal error estimates. In all cases, the key difference was the measurement of the main component of the fuel (in these tests, propene or propane). We discuss the challenges involved in these measurements.



INTRODUCTION Primary goals of the 2010 Comprehensive Flare Study included 1) assessment of the potential impact of high vent gas flow rate turndown on flare CE as well as on volatile organic compound (VOC) destruction and removal efficiency (DRE) and 2) determination of actual hydrocarbon DRE values under conditions commonly expected to yield DRE values of 98% or better. These tests also provided the opportunity to blind validate remote sensing techniques for CE measurements against extractive sampling techniques on full-scale industrial design flares. Aerodyne Research, Inc. (ARI) assembled an instrumentation suite to directly measure flare emissions in the flare exhaust plume and calculated DRE and CE based on those measurements. Several remote sensing technology developers also participated in the study and a comparison of their performance to extractive sampling was an important secondary goal of the study. Although several instrument developers and research groups participated in the tests, not all were directly comparable with the extractive sampling. The one group that did measure CE throughout the test series was Industrial Monitor and Control Corporation (IMACC), who operated Passive Fourier Transform Infrared (PFTIR) and Active Fourier Transform Infrared (AFTIR) instruments. This paper compares the CE values and underlying species measurements by ARI and IMACC, with the goal of assessing the capabilities and opportunities for further improvement in both techniques. © 2012 American Chemical Society

For the purpose of this study, combustion efficiency (%) is given by the ratio of gaseous concentrations in the flare plume as CE (%) = 100*[CO2 ]/([CO2 ] + [CO] + THCw )

(1)

where THCw is a weighted total hydrocarbon concentration, or a sum of hydrocarbon concentrations (molar or volume units) multiplied by the number of carbon atoms in each species. In the flares observed here, only a few species contributed significantly to THCw, as will be discussed below.



INSTRUMENTATION AND ANALYSIS Extractive Sampling. The ARI extractive sampling instrumentation suite is described in considerable detail in other papers in this issue.1,2 Here we only reinforce a few points needed in the discussion of the data sets. An important point to be made is that although the sampling instrumentation suite used involves several different measurement techniques, each can be relied upon to maintain accuracy while sampling combustion gases with their fluctuating mixtures of potentially interfering species. A reliable calibration is important, but in this application calibration with a known concentration of a pure species must be accompanied by specificity of detection in Special Issue: Industrial Flares Received: Revised: Accepted: Published: 12621

November 29, 2011 May 4, 2012 May 16, 2012 May 16, 2012 dx.doi.org/10.1021/ie202783m | Ind. Eng. Chem. Res. 2012, 51, 12621−12629

Industrial & Engineering Chemistry Research

Article

m3 per second of flare plume through the collector (this volume flow rate is within the range of variation of volume flow rates of total flare exhaust). Typical radial distances between collector inlet and flare exit ranged between 6 and 10 m. It can be seen from Figure 1 that the plume cross sectional area could be up to an order of magnitude larger than the area of the 50 cm diameter inlet cone of the sampling probe, so that typical fractions of the total exhaust plume drawn through the sample collector might be on the order of a few percent. A final point made in more detail in the accompanying papers1,2 is that the fluctuations inherent in measuring exhaust plume species as they mix with ambient air are dealt with, and in effect turned to advantage, by analyzing the linear relationships between species concentration data streams. The fact that all species concentrations have the same fluctuations due to changing production and dilution allows measurement of background concentrations as well as ratios of concentrations in the combustion exhaust, which are all that are needed to determine CE. The nominal error bars for the extractive sampling CE values were also generated as part of this analysis procedure: dividing the plot of propene against CO into two populations, above and below the fit line, and fitting each in turn, is the basis for an estimate of upper and lower bounds on CE that we will use here in comparison plots. These error limits are based on the observed random fluctuations within each test point data stream and have the advantages of being well-defined and capable of automated generation. They do not include estimates of possible systematic errors, although we will see below that they are not inconsistent with such estimates. Remote Sensing. The FTIR (Fourier transform infrared) remote sensing techniques used by IMACC were developed over many years.3−7 Current IMACC personnel developed a passive FTIR (PFTIR) instrument for a flare test at John Zink in 1984, funded by the US EPA Office of Research and Development in Research Triangle Park, NC.3 After that test, there was little interest in passive monitoring until the 2003 TCEQ test at John Zink.5 This was the first test that afforded an opportunity to challenge the methodologies. It also provided controlled data of a high quality so the algorithms and procedures could be tested and upgraded. The next significant test was in September of 2009 at Marathon, Texas City.6 IMACC and Clean Air Engineering approached this test believing more development would be required to perfect the passive technology. In reality, only minor alterations were needed in the algorithms the most significant changes were in calibration and test procedures. Between the Texas City test and the TCEQ 2010 Flare Study, several major flare programs were undertaken. These included the following: Shell - Deer Park, Texas February 20107 and Marathon − Detroit, Michigan July 2010.8 All of these programs added to the knowledge of the passive techniques and allowed for further refinement of the procedures. The instrumentation, procedures, and algorithms used in the TCEQ 2010 Flare Study were extensively modified and upgraded from that used in the 2003 TCEQ test. The improvements in data quality since that test reflect these developments. The basic concept of the IMACC PFTIR measurement is shown in Figure 2. The FTIR spectrometer collects emission spectra from the hot plume gas at the rate of roughly 1 spectrum per second. These spectra contain unique emission features for each gas present (indicated in Figures 3 and 4) whose intensities are proportional to the molecular species concentrations and the path length through the plume. Both

a mixture, either inherent in the technique or demonstrated by experiments. The CO2 concentration was measured using three Licor brand nondispersive infrared gas analyzers. The CO, methane, and ethene concentrations were measured using ARI quantum cascade laser (QCL) instruments based on tunable infrared laser differential absorption spectroscopy (TILDAS). These instruments display spectra, allowing immediate identification of any interfering species. As an example, propene absorption lines were noticed in the spectral region used for TILDAS measurement of ethene. These interfering lines were added to the real-time analysis using calibration gases, allowing the measurement of propene concentrations as well. However, although this data stream continued to agree well with the primary propene measurement, the large number of lines (typical for larger molecules) made accurate fitting more difficult and led to a noisier data set, which was not used in CE calculations. Inclusion of propene in the fit was, however, necessary to ensure that ethene was properly determined without interference. Propene was measured using proton transfer reaction mass spectrometry (PTR-MS), checked with calibration gas mixtures and validated against gas chromatography using a flame ionization detector (GC-FID). Of particular importance were experiments using a GC/PTR-MS system operated in parallel to the primary PTR-MS. Comparing chromatograms taken during calibration and during exhaust gas sampling, the propene mass signal was observed from a single chromatographic peak at the same retention time. This result confirms that the PTR-MS signal measured at that mass has only one source and that source is propene, so there are no other compounds present in the flare emission matrix that interfere with this measurement. A second point to be made is the considerable effort made to provide a sample collection system that brought adequate levels of the plume gas to the final sampling line. The sample collector, basically an 8 m length of 30 cm diameter pipe with an elbow and cone inlet, is seen in Figure 1. The eductor at the outlet end of the sample collector continuously drew up to 1

Figure 1. Flare plume sampling system (upper left) making measurements of flare plume from the steam assist flare while the sampling system is held in position by crane and ground crew. Behind the steam flare is the taller (unlit) air flare observed later in the tests. The scissors jack to the right of the crane holds the reflector for the active FTIR system. 12622

dx.doi.org/10.1021/ie202783m | Ind. Eng. Chem. Res. 2012, 51, 12621−12629

Industrial & Engineering Chemistry Research

Article

wavelength region is used to look at total organics in the C−H stretch region (3 μm) and at CO, CO2, NO, NO2 among other species. The longer wavelength region allows for speciation of individual organics as well as monitoring of CO2 at 16 μm. Because the signals observed with PFTIR are radiation emitted by the hot gases, the intensity of these emission features increases not only with concentration and path-length but also as the temperature of the emitting gas volume increases. The PFTIR instrument is typically operated at 0.5 cm−1 resolution, which is a compromise between signal-to-noise and high enough resolution to resolve the gas features of interest. The system telescope is a 30 cm diameter Cassegrain with an effective focal length of 483 cm. This has a divergence of less than 0.5 milliradians so the field of view at 100 m is not much larger than 0.3 m. Real time analysis converts these spectral feature intensities into species concentrations, using modified classical leastsquares (CLS) fitting routines in selected spectral regions. Initially, there was concern that gradients in the exhaust plume would have to be taken into account in the analysis. However, the field tests clearly showed that the plume being monitored consisted of pockets of hot gas embedded in an almost ambient air stream. The PFTIR was looking at a 0.3 m diameter area of the plume, and measurements were taken over fifteen minutes or more so the signal was averaged over the inhomogeneous plume providing a spatial and temporal average. As shown in the example in Figure 5, measurement of the resolved CO

Figure 2. Schematic of IMACC PFTIR measuring a flare plume. Observing the IR radiation emitted from the flare exhaust gases with a collection telescope and spectrometer allows individual species and gas temperature to be quantified. The field of view is aimed at a position approximately two flame lengths from the tip of the visible flame using a thermal imager. This is sufficiently downstream for combustion to be complete and gas temperatures to be moderate. At high CE values, the integrated gas temperatures and concentrations in the field of view are relatively constant, becoming more variable at low CE.

Figure 5. Plot of the log of intensity of CO lines in an observed flare spectrum versus J(J+1), a quantity proportional to initial quantumstate energy where J is the CO rotational quantum number. The slope of this curve is proportional to 1/T. In this case, the temperature was 497 K. (Taken from ref 6.)

Figure 3. Radiance spectrum showing the major emission regions of CO, CO2, and total organics (THC). The fingerprint region at smaller wavenumbers is a region where the individual organics have distinct features allowing each to be analyzed independently.

vibration−rotation lines also yields a temperature that is averaged over the entire emitting gas volume within the field of view. This set of average temperature and species concentrations is demonstrated to give acceptable values of CE. The fact that CE is a ratio means that the effective path length corresponding to the average temperature and species concentrations does not affect the CE determination. The nominal error bars generated as part of the IMACC automated analysis, used here in comparison plots, come from the residuals of the CLS fits in the various spectral regions assigned to individual species. The individual error estimates are then used in standard error propagation formulas to generate an uncertainty limit for CE. These individual species residuals from the CLS procedure are not indicative of random fluctuations in the observed FTIR spectra. They are indicative of how well the analysis procedures could identify and match all

Figure 4. Enlargement of the spectral region around 2200 cm−1 showing the CO and CO2 bands. The CO2 band has the center totally absorbed by the atmosphere making it look like two bands. The CO lines are used for temperature computation because the band shape changes with temperature.

the 3 to 5 μm (3000 to 2000 cm−1) atmospheric window and the 8 to 15 μm window (1250 to 650 cm−1) are used. The short 12623

dx.doi.org/10.1021/ie202783m | Ind. Eng. Chem. Res. 2012, 51, 12621−12629

Industrial & Engineering Chemistry Research

Article

paper came about through the desire of the measurement groups to look more closely at the data sets, including a first comparison between the PFTIR and Extractive Sampling data, to see what could be learned about the reliability of both methods and the opportunities for further technique development. We should point out that some aspects of remote techniques can only be developed through analysis of field data − “spiking” a combustion gas sample with calibration gases to identify interferences, a standard technique for sampling methods, is more difficult for a remote method like PFTIR.

spectral features present. If the choice of spectral regions and the fitting spectra at the disposal of the CLS procedure allow, the analysis can return a small residual, implying it fit the data well. This can be the case even when large random or systematic errors are indicated by comparison to other measurements. The calibrations performed with the PFTIR instrument using a hot cell and single reference gases spanning a range of concentrations show it to be capable of very accurate measurements.4 However, part of validating an analysis procedure is to minimize the residual errors when the gases of concern are in a matrix with possibly interfering compounds. The error bar returned when analyzing a complex matrix can demonstrate how well the routines were able to handle the matrix interferences. Prior to the 2010 TCEQ test, data from previous flare tests were used to refine the analysis methods and minimize the influence of interferences. One effect that can influence the data in an uncontrollable way is the inhomogeneity of the plume. This can cause a mismatch between the average spectral features in the plume and those represented by the reference set. The most significant influence is temperature. For this reason, the PFTIR field-of-view was held in a region where the plume temperature was as constant as possible and in the range of 250 to 350 °C. The reference set used was taken at 300 °C. IMACC also operated an Active Fourier Transform Infrared (AFTIR) instrument in the core steam tests. This technique uses an infrared emission source to perform an absorption measurement on the plume, thus freeing it from the limitation of detecting only gas at elevated temperatures. The infrared radiation from the source is modulated by the AFTIR, transmitted through the plume, and reflected back to the instrument using an array of corner-cube mirrors. This is a more straightforward measurement than PFTIR-radiance measurements, and it does not require the ancillary measurements that the PFTIR does. Because the AFTIR source is high temperature (1300 K), there was some expectation that this system would be better able to detect all organics and have better performance in minimum detection levels and data quality than the PFTIR. The AFTIR did not participate in the air flare tests as the air flare was taller, and the height limitations of the scissor lifts supporting the retro array would not allow proper alignment of the AFTIR, reflectors, and air flare plume. The AFTIR and PFTIR instruments provided similar CE values for all tests in which both sets of data were available. Because the PFTIR had a much larger data set we focus here only on PFTIR results. The goal of both the Extractive Sampling and the PFTIR measurement technique was automated, real-time analysis, not the generation of data sets that yield CE values only after months of tedious data reduction. As rehearsed above, the time required to develop this sophisticated level of hardware and software can stretch into decades, and every new field test can present new challenges, leading to failures in accurate measurement but also to opportunities to improve the measurements and analysis. The requirements of the 2010 Comprehensive Flare Study were that the measurement groups should provide first-look measurements on the spot and submit their final data reports on a time scale that precluded detailed, in-depth analysis and instead required automated analysis of the large data set. Further, IMACC submitted its data without access to the sampling measurements carried out by ARI. This provided a “blind” test which allows for independent comparison of the two methods. The review described in this



RESULTS Correlation between CE and Size of Error Estimates. Since the goal of this paper is comparison of the measurement techniques, not comprehensive presentation of results and discussion of their usefulness in flare design or compliance monitoring, we only present example data in Figures 6 and 7. In

Figure 6. Combustion efficiencies for steam flares and 80% propene/ 20% TNG and 100% propene fuels, as functions of combustion zone gas net heating value.

each figure, a substantial fraction of the complete data set is plotted against a parameter that is a measure of strength of combustion. The ratio of steam mass flow to vent gas mass flow, used in Figure 7, is self-explanatory, but the parameter in Figure 6 requires definition. The Combustion Zone Gas Net Heat Value (CZG NHV), Btu/scf, is defined as the ratio of the sum of the combustion heating value (LHV) of the vent gas going through the flare plus the combustion heating value of the flare pilots to the total volume of gases going to the flare, including steam or added air, i.e., vent gas plus pilot gas plus total steam or air assist. It is expressed numerically as CZG NHV ⎡ (VG) ⎣⎢ = ⎡ VG ⎢⎣ MWVG +

( )⎤⎦⎥ + ⎡⎣⎢(PG)( )⎤⎦⎥ ( ) ( ) + ( ) + ( )⎤⎥⎦ LHVVG MWVG

PG MWPG

LHVPG MWPG

S MWsteam

A MWair

(2)

where CZG NHV = combustion zone gas net heating value, Btu/scf; VG = vent gas mass flow rate, lb/h; LHVVG = vent gas lower heating value, Btu/scf; MWVG = vent gas molecular weight, lb/lb-mol; PG = pilot gas mass flow rate, lb/h; LHVPG 12624

dx.doi.org/10.1021/ie202783m | Ind. Eng. Chem. Res. 2012, 51, 12621−12629

Industrial & Engineering Chemistry Research

Article

Figure 8. Relative standard deviations computed for sets of test points in which all conditions were replicated.

estimates, we formed a “consensus fractional error limit” from the nominal error limits for Extractive Sampling and PFTIR so that CE values plotted versus this variable would retain the vertical correlation that allows us to compare the Extractive Sampling and PFTIR values in Figures 6 and 7. Figure 9

Figure 7. Combustion efficiencies for steam flares and 80% propene/ 20% TNG and 100% propene fuels, as functions of steam to vent gas ratio.

= pilot gas lower heating value, Btu/scf; MWPG = pilot gas molecular weight, lb/lb-mol; S = total steam mass flow rate, lb/ h; A = total air assist mass flow rate, lb/h; MWsteam = steam molecular weight, lb/lb-mol; MWair = air molecular weight, lb/ lb-mol. Figures 6 and 7 show the expected drop in CE as the fuel is diluted or as steam is added, and more importantly for our discussion, they show an increase in the nominal error bars for both techniques as CE decreases. These nominal error bars, introduced above, are for Extractive Sampling generated entirely based on observed fluctuations within a given test point data set. The PFTIR error bars, on the other hand, are representative of how well all features in the measured spectrum could be accounted for. However, the fact that a single measurement is over a volume with inhomogeneous concentrations, temperatures, and combustion efficiencies causes an increase in the error bars because no combination of reference spectra can exactly match this average of inhomogeneous spectra. Because of this basis in distortions introduced by inhomogeneities, we will continue to refer to these error estimates as based on random errors as well. A particularly welcome feature of the extensive CE data set is the good agreement between replicate measurements. Quite a few test points were repeated 2 or 3 times, often with substantial time separations. These observations often closely reproduce each other, as seen in Figure 8, which plots an array of relative standard deviations for these replicate sets, as a function of steam to vent gas ratio, a variable which we have already seen correlates inversely with CE, so the lowest CE values and largest standard deviations are found on the righthand side of the graph. The plot shows that the variances in both Extractive Sampling and PFTIR measurements increase as CE decreases. This implies that this is a real phenomenon, plume breakup, and not an artifact of the analysis procedures. In order to further reinforce the correlation between lower CE (leading to visibly greater fluctuations in flames and inhomogeneities in plumes) and observed random error

Figure 9. Defining a “consensus fractional error limit” by averaging the Extractive Sampling and PFTIR nominal error bars in quadrature and using this as an abscissa to plot the entire data set comprising steam flares with 80% propene/20% TNG and 100% propene fuels.

presents the entire steam flare/propene-containing fuels data set plotted against this variable, which is half the square root of the sum of the squares of the four error limits (2 differing limits for Extractive Sampling, 2 identical limits for PFTIR). This plot allows the reader to use the x-axis value to easily evaluate the nominal error bars. Figures 6, 7, and 9 show several general trends: the nominal error bars for high CE cases are often quite small, and those for low CE cases can be quite large. In the majority of cases the Extractive Sampling and PFTIR CE values agree to within their combined nominal error bars, but in some they do not. Granting that these nominal error bars are not the final word on expected error limits, we will turn our focus to those cases where they do not fully explain the disagreement between the two techniques and ask which component measurements are 12625

dx.doi.org/10.1021/ie202783m | Ind. Eng. Chem. Res. 2012, 51, 12621−12629

Industrial & Engineering Chemistry Research

Article

Figure 10. Ratios of the other major exhaust plume carbon-containing species to CO, as measured by extractive sampling (left) and remote PFTIR measurement (right). All concentrations used in the ratios are expressed as carbon equivalents, and the data set included both steam and air flares with 80% propene/20% methane fuels.

Figure 11. Plotting correlation between PFTIR and Extractive Sampling measurements of propene/CO ratios for all steam and air flares with 80% propene/20% methane or 100% propene fuels.

responsible for the disagreement and what this tells us about further technique development. The prime suspect for which component measurement is most critical in propene-fueled flares is revealed in Figure 10. The data plotted include both 100% propene tests and tests with the standard fuel of 80% propene/20% Tulsa natural gas, or TNG. Figure 10 shows that the ratio of propene carbon equivalents to CO is typically on the order of 10, while methane is approximately 10 times lower. Ethene (and, it turns out, the other hydrocarbons, several of which are measured either by Extractive Sampling or PFTIR or both) is lower yet, making it a relatively minor contributor. Figure 10 is also of interest in that it shows that all of these component ratios to CO are relatively smooth and relatively constant functions as CE changes, and in that the two versions of the figure based on Extractive Sampling and PFTIR data are seen to tell the same qualitative story. This is one major point to be made by this paper: both techniques are based on first-principles measurements of all the species (or ratios of species) needed to calculate an accurate value of CE. A detailed study of 176 test points confirms the above suspicion: significant disagreement between Extractive Sam-

pling and PFTIR CE values is always primarily attributable (for propene-fueled cases) to a difference in the amount of propene measured (relative to CO2, for example). Figure 11 provides a graphical representation of the situation. Here, the PFTIR error bars are the same, individual species residual errors described above, but the Extractive Sampling error bars are the total error bars for the individual species measurements as estimated a priori as part of the quality assurance work.4 These error estimates are not inconsistent with the test-point-observationbased nominal error bars on CE. It can be seen that quite a few test points in Figure 11 show agreement between Extractive Sampling and PFTIR measurements of propene/CO ratio, in that one or both error bars overlap the 1-to-1 line. Another cohort of points, not straightforwardly identified in this plot, belong to test points where CE was high, both propene and CO were low, and disagreements between Extractive Sampling and PFTIR measurements of either did not lead to disagreements in the resulting CE value. Finally, of the 141 test points represented in Figure 11, 58 fall into the class of differences in CE between Extractive Sampling and PFTIR exceeding the combined nominal error bars, and in all cases a reconciliation in 12626

dx.doi.org/10.1021/ie202783m | Ind. Eng. Chem. Res. 2012, 51, 12621−12629

Industrial & Engineering Chemistry Research

Article

Figure 12. Correlation between propene and methane for the 80% propene/20% TNG fuel data sets (steam and air flares) for Extractive Sampling and PFTIR.

decomposition and/or combustion of propene were just a bit more likely than that of methane, a plausible expectation. The PFTIR measurements show a similar correlation but a substantially larger variation. Again, we can comment that least-squares fitting of mixture spectra is an art that can be perfected only through extensive testing, and the results in Figure 12 demonstrate both qualitative success in the difficult task of identifying propene spectral features in combustion gas emission spectra and difficulties in quantifying propene that can in some cases lead to significant errors in CE. Examination of correlations between other pairs of species yielded some expected results. For instance, CO and propene are clearly less correlated, as expected given their different sources. On the other hand, the correlations between ethene and CO in the data of both measurement groups are very good indeed, with PFTIR data sets again showing more scatter. A substantial subset of the test series used propane rather than propene as the fuel. Neither measurement group was fully prepared for this development, as this fuel was added after the test was underway. ARI replaced the PTR-MS measurement of propene with a flame ionization detector (FID) measurement of total hydrocarbon in the sampled exhaust, which in some cases was verified by a total carbon determination made using an oxidation catalyst and a LiCor CO2 monitor. The PFTIR CLS fit for propane did not yield significant amounts even when the extractive sampling indicated large amounts, and the flare conditions were equivalent to those propene-fueled flares that yielded low CE values. We conclude that the propane cases are not useful for our present purposes of comparing the two techniques at their current best.

propene measurement (relative to CO2, for instance) would bring them into agreement. Cases with Disagreements Larger than Nominal Error Bars. Figure 11 does not provide any suggestion as to which error bars, Extractive Sampling and/or PFTIR, ought to be extended to result in CE error estimates that explain all differences. We have considered extractive line-losses and issues involving the PFTIR analysis. We know that the PFTIR reference set available for C2H4 and C3H8 at the time of the TCEQ test was limited. For this test, the reference set was compiled from the best elevated temperature spectra then available. The references for C2H4 and C3H6 were from the IMACC 185 °C library. Subsequent to the TCEQ test, the US EPA funded development of families of high temperature spectra for CO, CO2, CH4, C2H4, C3H6, and H2O. These were laboratory generated at a temperature of 250 °C using a 0.5 m path length and concentrations as high as 5000 ppm. A review of the quality assurance procedures for the extractive sampling and analysis techniques4 suggests that systematic (bias) errors from this procedure should be minimal. However, a sample recovery study was not conducted as a part of this investigation. Direct checks on the CLS fitting procedure involved in the PFTIR measurements, equivalent to the extractive sampling quality assurance assessment, are harder to perform. Hot cell calibration tests have been done on CO and CO2 but are yet to be done on C2H4, C3H6, CH4, and water at 250 °C. Hot cell tests demonstrate the accuracy and precision of the analyses in the absence of interfering compounds. Additional tests of this type using gas mixtures representative of all plume constituents would be valuable. In addition, reprocessing selected spectra with a method including the new high temperature references, discussed above, could determine if reference scaling error is significant. In advance of that work, we can look to the data to see which propene measurements correlate well with other measurements and what this says about the random errors in the Extractive Sampling and PFTIR data sets. The species measurement we would expect to show best correlation with propene is methane, for the subset of flares with 80% propene/20% methane fuel. We speculated above that the major source of both of these species in the exhaust plume is pass-through of unburned fuel, while all the other carbon-containing species are products of combustion. In Figure 12 we see that the Extractive Sampling measurements of propene and methane do indeed correlate well, with the amount of propene typically being a bit less than the 4-to-1 ratio in the input fuel − as if the thermal



DISCUSSION We began this review of the gas phase measurements in the 2010 Comprehensive Flare Study with two concerns regarding the limits to accurate measurement: that the greater fluctuations in component signals with lower CE made accurate measurement difficult and that plume inhomogeneity might limit the accuracy of techniques that sample only small portions of the plume. These issues do indeed pose challenges to accurate measurement, but the evidence in the actual data sets suggests that our fears should be tempered and that technique improvements with the most potential to improve accuracy may be in other areas. Two qualifying statements, each with discussion, follow. 12627

dx.doi.org/10.1021/ie202783m | Ind. Eng. Chem. Res. 2012, 51, 12621−12629

Industrial & Engineering Chemistry Research

Article

Fluctuations in Component Measured Quantities Can Be Much Larger than Fluctuations in CE. The foundation for this observation is discussed in detail in accompanying papers1,2 that demonstrate that the various exhaust species all mix with the ambient at very close to the same rate, so that correlating these species can lead to very accurate values of the ratios which are all that calculation of CE requires. The evidence presented here that this is in fact true includes the low relative standard deviations in replicate observations (mostly under 5%) seen in Figure 8 and the fact that a number of the extractive sampling points in Figure 9 with CE below 30% still have nominal error bars in the range of 10%. Inhomogeneities in the Visible Flame Do Not Imply Similar Inhomogeneities in the Plume. The flame seen in Figure 1, far from being a single, compact, completely connected volume, is clearly capable of breaking into freely floating volumes of burning gas separated from each other by nonluminous gas. This naturally led to a mental model in which as CE decreases and this inhomogeneity in the visible flame increases, the exhaust plume beyond the flame also contains more and larger volumes of unburned gas, which can then mix with the atmosphere and escape combustion entirely. Our particular concern was that while the sampling, carried out through a probe that had been demonstrated to thoroughly mix all sampled gas, would not be directly affected by this, the remote methods could be very seriously affected, depending on which type of gas volume was in their field of view or line of sight. However, the data set from the PFTIR does not bear out this fear, at least in its most extreme form. We have to remember that while the still-burning volumes in an inhomogeneous visible flame are by definition unmixed, the transport processes of mass and heat are proceeding in concert, and when the visible flame is gone, what remains is warm, wellmixed combustion gas, which then proceeds to mix with ambient air. The PFTIR detects thermal emission from warm gas, whether it be the products of combustion such as CO2, CO, or C2H4 or the unburnt fuel species such as propene and methane. Although in principle propene and methane could be products of fuel decomposition as well, the fact that Figure 12 shows propene/methane ratios close to that in the input fuel suggests that most of the propene and methane in the exhaust plume simply passed through unreacted. However, if propene arrived in the exhaust plume only through unburned regions of cold gas, it would not register a PFTIR signal. Figure 11 shows this is not the case. Indeed, if propene, the most significant carbon-containing species after CO2, were partially missing from PFTIR observations, we would expect PFTIR CE values to be systematically high. Reviewing 33 steam flare/propene fuel cases with Extractive Sampling/PFTIR disagreements larger than the nominal error bars, 14 PFTIR CE values are larger than the extractive sampling values, and 19 are lower. On the other hand, of 25 comparable air flare cases, 23 PFTIR CE values are lower than Extractive Sampling (the ratio of PFTIR propene/CO2 to Extractive Sampling for these cases is 2.4 ± 0.9). The difference could involve different levels of inhomogeneity in the two plumes, but it also could involve differences in the FTIR spectra: additional water interference in steam flare spectra could mean that an analysis algorithm that found, on average, about the right level of propene, could when applied to air flares overestimate the propene levels. This brings us to the point of the discussion: while reducing the effects of inhomogeneities on remote sensing involves increasing averaging time or field of view size, each with its own negatives

for efficient measurement, further examination of the details of the CLS fitting of FTIR spectra may yield gains in analysis reliability which substantially improve performance.



CONCLUSIONS The 2010 Comprehensive Flare Study involved the first blind validation of a remote sensing technique for flare combustion efficiency against extractive analysis techniques. The additional analysis reported here focused on individual species measurements and on cases where CE values measured by the two techniques differed by more than the nominal error estimates. The results presented here include the following: • Both techniques were shown to have a sound basis in first principles measurements of the components of CE, capable of demonstrating expected trends in ratios of the several combustion species measured. • In all cases where the two techniques differed significantly, the key difference was the measurement of the main component of the fuel (in these tests, propene or propane). • The fluctuations and inhomogeneities in flare exhaust plumes are real problems for CE measurement; however, they do not pose limitations that are as serious as expected. • Further technique development is possible, and future comparison tests should show even better agreement between extractive sampling and remote sensing of flare combustion efficiency and related parameters. • We conclude that when all major species are accurately measured, both techniques can be profitably applied to a variety of combustion systems, including ones with much greater fluctuations and inhomogeneities than a properly operating flare.



AUTHOR INFORMATION

Corresponding Author

*Phone: 978-663-9500, x229. Fax: 978-663-4918. E-mail: [email protected]. Present Address §

Department of Public Health, University of Massachusetts, Amherst MA. Notes

The authors declare the following competing financial interest(s): Aerodyne Research, Inc. and Industrial Monitor and Control Corp. sell products and services related to the measurement techniques described in the paper, while the remaining co-authors have participated, and may again participate, in measurement programs involving ARI and/or IMACC.



ACKNOWLEDGMENTS This work was supported by the State of Texas through the Air Quality Research Program administered by the University of Texas at Austin by means of a grant from the Texas Commission on Environmental Quality.



REFERENCES

(1) Knighton, W. B.; Herndon, S. C.; Franklin, J. F.; Wood, E. C.; Wormhoudt, J.; Brooks, W.; Fortner, E. Direct measurement of volatile organic compound emissions from industrial flares using real-time online techniques: proton transfer reaction mass spectrometry and 12628

dx.doi.org/10.1021/ie202783m | Ind. Eng. Chem. Res. 2012, 51, 12621−12629

Industrial & Engineering Chemistry Research

Article

tunable infrared laser differential absorption spectroscopy. Ind. Eng. Chem. Res.2012, 51, doi.org/10.1021/ie202695v. (2) Herndon, S. C.; Nelson, D. D.; Wood, E. C.; Knighton, W. B.; Kolb, C. E.; Kodesh, Z.; Nettles, R.; Torres, V. M.; Allen, D. T. Using the principle of carbon-balance to quantify flare emissions. Ind. Eng. Chem. Res. 2012, 51, doi.org/10.1021/ie202676b. (3) Spellicy, R. L.; Persky, M. J. Flare effiency monitoring by remote sensing. In Infrared methods for gaseous measurements: theory and practice; Wormhoudt, J., Ed.; Dekker: New York, 1985; p 139. (4) Allen, D. T.; Torres, V. M. TCEQ 2010 Flare Study Final Report; The University of Texas at Austin, 2011. Available http://www.tceq. texas.gov/airquality/stationary-rules/flare_stakeholder.html. (5) URS Corporation, Passive FTIR Phase I Testing of Simulated and Controlled Flare Systems: Final Report; Prepared for Texas Commission on Environmental Quality, June 2004. Available http:// www.tceq.texas.gov/assets/public/implementation/air/am/contracts/ reports/oth/Passive_FTIR_PhaseI_Flare_Testing_r.pdf. (6) Marathon Petroleum Company, LLC, Texas Refining Division. Performance Test of a Steam-Assisted Elevated Flare with Passive FTIR, Final Report; Texas City, Texas, May, 2010. Available http:// www.tceq.texas.gov/assets/public/implementation/air/rules/Flare/ 2010flarestudy/mpc-txc.pdf. (7) Shell Global Solutions (US) Inc., Shell Deer Park Refining LP Deer Park Refinery East Property Flare Performance Test Report; 2011. Available http://www.tceq.texas.gov/assets/public/ implementation/air/rules/Flare/2010flarestudy/sdp-epf-test.pdf. (8) Marathon Petroleum Company, LP, Detroit Refinery. Performance Test of a Steam-Assisted Elevated Flare with Passive FTIR − Detroit, Final Report; Detroit, Michigan, November 2010. Available http://www.tceq.texas.gov/assets/public/implementation/air/rules/ Flare/2010flarestudy/mpc-detroit.pdf.

12629

dx.doi.org/10.1021/ie202783m | Ind. Eng. Chem. Res. 2012, 51, 12621−12629