United States Acid Rain Division EPA/430-R-99-009a iS
Environmental Protection (6204J) June 1999
Agency Washington, D.C. 20460
&EPA EPA Flow Reference Method
Testing and Analysis:
Findings Report
-------
EPA FLOW REFERENCE METHOD
TESTING AND ANALYSIS
FINDINGS REPORT
Preparedfor
U.S. Environmental Protection Agency
Acid Rain Division, 6204J
401 M Street, SW
Washington, DC 20460
Contract No. 68-D7-0061
Task Order No. 02
U.S. Environmental Protection Agency
Region 5, Library (PL-12J)
77 West Jackson Boulevard, 12th floor
Chicago, IL 60604-3590
Prepared by
The Cadmus Group, Inc.
1920 Highway 54, Suite 100
Durham, NC 27713
June 1999
-------
EXECUTIVE SUMMARY
In the summer of 1997, the U.S. Environmental Protection Agency (EPA) and the Cadmus Group,
Inc. (Cadmus) conducted a series of week-long field tests at three electric utility sites to evaluate
potential improvements to the Agency's reference method for measuring volumetric flow in stacks
(Test Method 2,40 CFR Part 60, Appendix B). Method 2 does not include procedures to account
for yaw or pitch angles of flow when the flow in the stack is not axial and allows the use of only two
probe types, the Type S and the Prandtl. The three test sites were chosen to provide a range of yaw-
and pitch-angle flow conditions. After undergoing extensive wind tunnel evaluation, seven types
of velocity sensing probes (Prandtl, manual Type S, automated Type S, French, modified Kiel, DAT,
and spherical) were included in field tests. The manual and automated versions of the Type S probe
were operated with and without yaw angle determination. The resulting nine probe/procedure modes
were evaluated for their field worthiness and potential to improve the accuracy and precision of
Method 2.
The experimental design for the field tests was structured to yield data that would allow drawing
statistically and scientifically supportable conclusions about inter- and intraprobe performance, test
team effects, and the impact of velocity decay near stack walls on volumetric flow rate
measurements. Four test teams using multiple copies of each tested probe performed the field tests.
The in-stack measurements were compared to estimates of volumetric flow obtained using five
engineering methods: four based on boiler stoichiometry and one based on boiler efficiency.
Extensive measures were taken before, during, and after the field tests to ensure data quality. Two
rounds of data analyses were performed: one in preparation for peer review, another in response to
peer review. A battery of analytical and statistical tools was used to assess the accuracy and
variability of the probes and procedures.
The analysis resulted in the following key findings.
• Using various measures of accuracy, three probes yielded the most favorable results: the
automated and manual Type S yaw-nulled probes, and the DAT probes. Excluding the low-
load tests at one site (which presented difficulties for all probes except the automated Type
S), across all three test sites, the analysis revealed that all three probe types produced velocity
values within ±3% of the mean of all methods and the engineering baseline.
• Of all the tested probes, the automated Type S, in either straight-up or yaw-nulled mode,
yielded the most consistent results. Across all test sites and all load conditions, its
coefficient of variation when performing a 16-point traverse was always below 1%.
• The spherical, modified Kiel, and French probes yielded the most variable results with
coefficients of variation as high as 19.7%, 5.2%, and 4.2%, respectively, and deviations from
the first-round grand mean as large as 12.3%, 7.3%, and 3.3%. A number of peer reviewers,
however, questioned the plausibility of the spherical probe data obtained in the second field
test. If the spherical probe results from the second field test are not included, the probe's
accuracy and variability results are comparable to the accuracy and variability of the
automated and manual Type S yaw-nulled probes and the DAT probe. Velocity values
-------
produced by the spherical probe were within ±2.3% of the mean of all methods and ±2.4%
of the engineering baseline. Its coefficient of variation did not exceed 1.4%.
• A procedure to account for velocity decay near the stack wall was used during the three field
tests and in follow-up testing at additional sites. Accounting for "wall effects" resulted in
volumetric flow reductions ranging from 0.6% to 2.4% after excluding anomalous results.
The average percent difference in velocity due to wall effects was -1.5% for steel stacks and
-1.9% for brick and mortar stacks.
The above results suggest that disparities that have sometimes been reported between engineering
estimates of flow and in-stack measurements using Method 2 can be eliminated by using (1) probes
that can reliably account for the yaw and pitch angles of flow (e.g., manual and automated Type S
yaw-nulled, DAT, and spherical probes); and (2) a procedure that can account for velocity decay near
the stack wall.
IV
-------
OVERVIEW
PURPOSE OF THE STUDY
This report describes an experimental program sponsored by the U.S. Environmental Protection
Agency (EPA) to evaluate potential improvements to the Agency's current reference method for
measuring volumetric flow (Method 2,40 CFR Part 60, Appendix B). Method 2 [Determination of
Stack Gas Velocity and Volumetric Flow Rate (Type S Pitot Tube)] specifies measurements to
determine volumetric flow, but does not prescribe specific procedures to account for yaw or pitch
angles of flow when the flow in the stack is not axial. Method 2 also allows the use of only two
probe types, the Type S and the Prandtl.
During the summer of 1997, EPA conducted week-long field tests at three electric utility sites to
evaluate potential improvements to Method 2. The candidate improvements grew out of a technical
review of Method 2 and draft Method 2F (an unpromulgated, mid-1993 proposal for using three-
dimensional probes in situations where flow velocity has significant yaw and pitch components),
extensive preliminary wind tunnel and field testing, and comments provided by industry on the
performance of currently available technology.
Three test sites were chosen to provide a range of flow conditions under which the tested equipment
and procedures could be rigorously evaluated. One site had near-axial flow, another had flow with
moderate yaw and pitch angles, and the third had flow with a significant yaw component. Two sites
were gas-fired and the third was coal-fired. This provided an opportunity to compare in-stack
measurements of gas flow with calculations of volumetric flow using engineering methods for more
than one type of fuel. "Wall effects" tests were conducted at the three primary sites and six
additional stacks to measure the impact of velocity decay near the stack wall on volumetric flow.
The four primary goals of the field study were to:
• identify changes to Method 2 that could improve the accuracy of the method under
various flue gas flow conditions,
• ensure that each contemplated change would result in measurements that have acceptably
low variability,
• ensure that each change would be practicable for use in the field and able to withstand
the conditions typically encountered in electric utility stacks, and
• ensure that any changes to the methods would not result in harm to the environment
because of systematically low measurements of volumetric flow.
Two collateral goals were to (1) seek improvements to Method 2, which might reduce the disparity
industry had reported to occur under certain circumstances between the heat rate derived from in-
stack measurements of flow and heat rate calculated from combustion parameters; and (2) collect
data that would quantify wall effects.
-------
FIELD STUDY PROCEDURE
Tests compared flow results for seven different types of probes, some operated in different modes,
for a total of nine probe/procedure combinations. The nine combinations were the manual Type S
probe straight-up (the current Method 2 procedure), manual Type S probe yaw-nulled, automated
Type S probe straight-up, automated Type S probe yaw-nulled, DAT, spherical, modified Kiel,
French, and Prandtl. Probe types were initially qualified for inclusion in the study through testing
performed during 1996 at the Merrill Subsonic Wind Tunnel at North Carolina State University
(NCSU) in Raleigh. All of the probes used in the field tests were provided to the test teams by EPA,
except for the DAT probes, which the test teams supplied.
Four stack-testing teams performed simultaneous measurements using Method 2 and pre-selected
alternatives to Method 2. The multiple-team approach ensured that the data collected on each tested
alternative would be representative of and support an assessment of the extent and sources of
variability in the measurements. Concurrent in-stack measurements were made using the automated
Type S system (the Autoprobe) and using a cross-correlation infrared monitor. The cross-correlation
monitor was evaluated as an adjunct to this study. Simultaneously with the in-stack measurements,
plant parameter data were collected so that volumetric flow could be estimated using the engineering
methods.
A three-part experimental design was used in the field tests:
1. Matrix A—In this interprobe comparison, simultaneous measurements from different types
of probes were compared to examine the relative magnitudes of volumetric flow
measurements determined by different probe types.
2. Matrix B—In this intraprobe comparison, multiple copies of the same types of probes were
compared. In conjunction with Matrix A, Matrix B was designed to provide more extensive
data on the variability in each probe's measurements.
3. Matrix C—The wall-effects study prescribed the collection of data in 18 1-in. increments
from the stack wall. This study was conducted to evaluate the extent of decay in flue gas
velocity near the stack wall.
QUALITY ASSURANCE
Extensive measures were taken to ensure data quality, the most important of which were probe
calibrations and pre-analysis screening of data.
Probe Calibration. Routine calibrations of all copies of all probes were conducted at NCSU before
the start of field testing (designated as pre-test calibrations) and after all testing was completed
(designated as post-test calibrations). Intermediate calibrations were performed for some probes
between the first and second field tests because the probes were physically changed. Field test data
were derived using the most recent calibration performed before each field test.
The NCSU post-test calibrations were used to determine if the probe calibrations changed over the
course of testing and the extent to which any changes affected test results. Independent calibrations
of the test probes were performed by the National Institute of Standards and Technology (NIST) at
VI
-------
its Dual-Test Section Wind Tunnel in Gaithersburg, Maryland, using the same written protocol used
during the NCSU calibrations.
Data Screening. A three-step, systematic procedure was implemented to identify individual
volumetric flow values or specific methods that should not be included in data analyses due to data
quality flaws or to problems in the operation of the method.
In the first step of the data screening process, plots of volumetric flow versus run were examined to
identify measurements that appeared to differ substantially from other values in the same or proximal
runs. Raw data records and supporting information associated with all values thus identified were
then reviewed. Data quality problems, equipment malfunctions, and procedural problems were
noted. Correctable errors identified by reviewing data records and supporting information were
corrected. Values that were flawed due to documented procedural or equipment problems were
excluded from analysis.
After this initial screening was completed, the interquartile range (IQR) was used to identify
additional suspicious values. The IQR is the difference between the 25th percentile value and the
75th percentile value of all measurements taken for a specific method. Values falling outside 1.5
times the IQR were flagged, and the raw data and supporting information were reviewed.
Finally, the stability of each unit's operation during the course of the field test was evaluated by
plotting volumetric flow, O2 concentration at the economizer outlet, and unit operating load for each
run for each site. The resulting effect of unit instability on measured volumetric flow values was
examined.
INITIAL DATA ANALYSES
Comparison of Volumetric Flow Measurements Across Probes and Methods. The Matrix A
(interprobe) and Matrix B (intraprobe) test results were used to examine differences across probe
types and methods. One round of field data analysis was conducted before the results were peer
reviewed. A second round of analysis (discussed below) was conducted that incorporated the
recommendations provided to EPA by the peer reviewers. Method comparisons were made using
several analytical approaches, including descriptive statistics and graphs to describe the flow
characteristics at each site and to compare the volumetric flow results obtained by the various
methods; rank order analyses to investigate patterns among the methods; analysis of variance
(ANOVA) on the Matrix A volumetric flow values to determine whether statistically significant
differences can be detected among the individual methods and classes of methods; and central
tendency analyses to determine whether certain methods are more likely to be good indicators of the
average, long-term volumetric flow.
Each in-stack method was examined individually to determine whether the choice of particular probe
copies and test teams affected the results obtained. The variability of each method was also
examined. The expected uncertainty in engineering methods was compared with the expected
uncertainty in the probe methods.
Analysis of Wall Effects. Wall effects tests were conducted at each of the three steel-stack field test
sites during the day under steady load conditions (Matrix C runs) using manual probes and
-------
Autoprobes, and at night under nonsteady-state load conditions using only Autoprobes. Additional
tests using only Autoprobes were conducted at six sites with brick and mortar stacks to investigate
the effect on volumetric flow of velocity decay for rough stack surfaces.
A standard 12-, 16-, or 20-point cross-stack traverse was first performed as prescribed in Method 1
and the average stack gas velocity was calculated. Velocity was then measured at 1-in. increments
across the width of each of the four Method 1 equal-area sectors adjacent to the stack wall, usually
starting at 1 in. from the stack wall and moving toward the center of the stack. The near-wall
measurements were used to calculate a replacement velocity value for the equal-area sector adjacent
to the stack wall for each of the four sampling ports. The average cross-stack velocity was then
recalculated by replacing the measured velocity values for the four equal-area sectors closest to the
stack wall with the calculated replacement velocity values. The percent difference between the
original average cross-stack velocity and the average cross-stack velocity making use of replacement
velocity values was then calculated. The calculated percent difference indicates the effect on
volumetric flow due to velocity drop-off close to the stack wall. The variation of velocity drop-off
as a function of stack material was examined.
SUPPLEMENTAL DATA ANALYSES
The second round of data analyses included a refinement of the central tendency analysis, the
addition of a comparison of probe measurements to an engineering baseline, and the performance
of a sensitivity analysis of the effect of excluding certain data points that appeared implausible to
peer reviewers, even though no documented problems could be found to justify their exclusion based
on the criteria established for identifying outliers. A comparison of pressure data collected by
manual and electronic devices at two wind tunnels and several field test sites was also conducted.
The concerns expressed by the peer reviewers regarding the Round 1 central tendency analysis
centered on how the grand mean was derived. Reviewers indicated that calculating the grand mean
of the volumetric flow values using the measurements obtained by every probe type used in the study
may not be appropriate, for several reasons. Some probe types used in the study produced uniformly
high or highly variable results. Some probe types and methods were over-represented in the grand
mean calculations (e.g., there were four Type S probes, each of which was operated straight-up and
yaw-nulled and five engineering methods). Because probes or methods of the same type tend to
produce similar flow measurements, the Round 1 grand mean value reflects the over-represented
probe types or methods, and therefore may not be a good estimate of the true flow in the stack.
Probes that do not account for yaw angles generally give higher readings than those that do and
should not be included in calculations of the grand mean. And finally, because velocity decay near
the stack wall is not adequately captured by a standard Method 1 sampling traverse, the volumetric
flow values used in the Round 1 grand mean calculations do not account for wall effects and are
therefore high.
Based on these concerns and suggestions, a second central tendency analysis was performed,
calculating the grand mean using only those methods reviewers considered to be most technically
credible and using wall effects-adjusted values, based on data derived during the wall effects runs.
Additionally, Run 3 for the spherical probe at Lake Hubbard was excluded because it seemed
implausible to several peer reviewers.
-------
The refined grand mean was calculated using flow measurements from the following four probe
types and two engineering methods: manual Type S yaw-nulled, DAT, spherical, baseline
Autoprobes 16-point yaw-nulled, MMBtu, and BTCE. For DeCordova and Lake Hubbard, the
Autoprobes 16-point yaw-nulled, DAT probes, MMBtu, and BTCE each contributed eight flow
values to the grand mean, while the Type S yaw-nulled and spherical probes each contributed four
flow values. For Homer City, each probe type and method contributed four values to the grand
mean, except for the spherical probes, which contributed eight.
KEY FINDINGS
DAT Probe. The DAT probe gave favorable results with respect to approaching the central tendency
of the data (within ±1.0% of the grand mean at DeCordova, Lake Hubbard—high load, and Homer
City and 2.3% below the grand mean at Lake Hubbard—low load). This finding is supported by the
refined central tendency analysis in Section 4 (DAT probes were always within ±1.3% of the grand
mean across all sites). The DAT probe's variability was typical of manual probes (the coefficient
of variation ranged from 2.08% at Homer City to 3.14% for the low-load runs at Lake Hubbard), but
was significantly higher than that typical of automated probes. The ANOVA analyses of volumetric
flow indicate that variation in flow is attributable to different copies of the probe and, at one site, test
team was a significant factor in flow variation.
Prandtl Probe. Although the coefficient of variation of the Prandtl probe at the two gas sites where
it was tested was the lowest of any of the manual probes (0.63% and 1.27%), mean flow readings
at both sites were higher than the central tendency. The Prandtl probe had the highest deviation from
the central tendency (+4.3%) at the near-axial site (DeCordova); the deviation was less severe
(+2.2%) at the moderate yaw/moderate pitch site (Lake Hubbard). This finding is consistent with
the refined central tendency analysis in Section 4 (3.53% from the central tendency at DeCordova
and 2.08% at Lake Hubbard).
Spherical Probe. The four original spherical probes (damaged at DeCordova and subsequently
repaired and re-calibrated before Lake Hubbard) produced moderately high flow values at
DeCordova and by far the lowest flow values for both the high- and low-load tests at Lake Hubbard
(12.3% and 10.6% below the grand mean). If Run 3 is excluded, as suggested by peer-reviewers,
the average flow measurement for spherical probes for Lake Hubbard high-load tests is lower than
the grand mean by 3.45%. At Lake Hubbard low-load, the percent deviation from the grand mean
was three times greater than that of any other method. On the other hand, at Homer City where both
the new and old sets of spherical probes were tested, the volumetric flow values were very close to
each other and much closer to the central tendency of the data (2.0% and 2.1% below the grand
mean). The rank order of the spherical probes was relatively high at DeCordova, lowest at Lake
Hubbard, and in the middle at Homer City. Consistent results were observed from the refined central
tendency analysis in Section 4, except that the deviation from the central tendency was smaller across
all sites.
Autoprobes Yaw-Nulled. Of all the tested probe types, the baseline Type S Autoprobe system
operated in the yaw-nulled mode with 16-point traverses was closest to the central tendency across
all three sites and load levels, as found in both the original and refined central tendency analyses
(0.3%, -0.1 %, -0.4%, and -1.6% for DeCordova, Lake Hubbard high- and low-load, and Homer City,
respectively, from the central tendency in the original analysis, and -0.23%, -0.20%, 0.61%, and
IX
-------
-0.53% for DeCordova, Lake Hubbard high- and low-load, and Homer City, respectively, from the
central tendency in the refined analysis). Results for 12-, 16-, and 48-point traverses for the baseline
Autoprobe system and the 16-point traverses for the single manual Autoprobe were comparable. The
12-, 16-, and 48-point baseline Autoprobes yaw-nulled and the manual Autoprobe yaw-nulled
displayed the lowest variability among all the methods tested, with coefficients of variation
consistently below 1%.
Autoprobes Straight-Up. Straight-up operation of the Autoprobes consistently produced higher flow
values than yaw-nulled operation. As expected, the difference between the straight-up and yaw-
nulled modes increased as flow angularity increased from near-axial to moderate yaw/moderate pitch
to high-yaw flow conditions. For 16-point traverses, the deviation from the grand mean for straight-
up operation increased from 0.3% at DeCordova, to 1.2% and 0.9% for Lake Hubbard high- and low-
load, to 2.7% for Homer City. At the near-axial site (DeCordova) the deviation from the grand mean
for straight-up operation (for 16-point, 48-point, or manual 16-point) was no more than 0.2% larger
than the deviation from the grand mean for yaw-nulled operation. At the other sites, differences
between the straight-up and yaw-nulled values were substantially larger (up to 1.4% at Lake Hubbard
and 4.4% at Homer City). The low variability observed for yaw-nulled operation was also found for
straight-up operation.
Type S Probe Yaw-Nulled. The Type S probe operated in the yaw-nulled mode consistently
produced flow results 2.2% to 2.9% higher than the grand mean. The results of the refined central
tendency analysis in Section 4 are consistent with these results. The coefficients of variation for the
Type S probe yaw-nulled across all three sites ranged from 1.76% to 3.23%, which was typical for
the manual probes tested. No significant differences among probe copies were found, but at Lake
Hubbard some test team effect on flow measurement was detected. As with the Autoprobes, yaw-
nulled operation consistently produced lower flow values than straight-up operation, and the
difference between the two modes increased as flow angularity increased from near-axial to
moderate yaw/moderate pitch to high-yaw flow conditions.
Type S Probe Straight-Up. The Type S probe operated in the straight-up mode has been the
standard method for measuring volumetric flow for over 20 years. The Type S probe straight-up had
the highest positive difference (6.9%) from the central tendency at Homer City and in low-load
operation at Lake Hubbard (3.4%), second highest for high-load at Lake Hubbard (6.0%), and third
highest at DeCordova (3.5%). In the refined central tendency analysis in Section 4, Type S probe
straight-up had the highest positive deviation from the central tendency at Lake Hubbard high- and
low-load and Homer City (5.69%, 4.15%, and 7.62%, respectively), and the second highest at
DeCordova (2.52%). At the same time, its measurements at the near-axial site (DeCordova) were
slightly lower than those of the Prandtl, which is the accepted standard pitot used to calibrate other
probes in wind tunnels with axial flow. The coefficient of variation of the Type S straight-up
measurements was 2.31% or less at all three sites.
French Probe. At the moderate yaw/moderate pitch and high-yaw angle sites (Lake Hubbard and
Homer City), the French probe produced volumetric flow values that were lower than the central
tendency (-2.0% and -3.3%). At Homer City the French probe produced the lowest measurements
of all the tested in-stack methods (3.3% below the grand mean). The French probe's coefficient of
variation ranged from 1.62% at the gas-fired site with axial flow (DeCordova) to 4.18% at the coal-
-------
fired site, where it was the most variable of all tested methods. Its rank order relative to the grand
mean, which varied from +6 at DeCordova to -6 at Lake Hubbard and -8 at Homer City, suggests
a strong dependence of flow measurement capability on yaw angle.
Modified Kiel Probe. The modified Kiel probe produced the highest volumetric flow values of all
the tested methods at Lake Hubbard high-load (7.3% above the grand mean) and second highest at
DeCordova (3.6% above the grand mean). However, at Homer City, the modified Kiel probe had
the second smallest deviation from the grand mean (0.3%) of all tested methods. At Lake Hubbard
low-load, it had the smallest deviation (2.0%) from the grand mean of all the tested in-stack manual
methods. The modified Kiel probe's coefficient of variation at Homer City (1.24%) was lowest of
any of the tested manual probes. At DeCordova, the coefficient of variation (1.39%) was in the
middle of the range for manual probes. The probe had the second highest coefficient of variation
at Lake Hubbard (5.19% at high-load and 4.04% at low-load).
Comparison to Engineering Baseline. At DeCordova, Lake Hubbard high- and low-load, and
Homer City, respectively, the following percent differences from the MMBtu method flow values
were found: 0.39%, -1.88%, -3.55%, and -1.44% for the Autoprobes 16-point yaw-nulled; 2.45%,
0.04%, -1.16%, and 1.88% for the Type S probe yaw-nulled; and 0.37%, -2.71%, -5.33%, and
-0.86% for the DAT probe.
Comparison of Pressure-Measuring Devices. From a practical standpoint, differences in pressure
readings between manual and electronic pressure devices were generally small. Where statistically
significant differences were detected, the differences were not consistent for different types of probes
and different velocity conditions. When used with Type S probes, however, manual devices tend
to read higher pressures than electronic transducers in field tests, but read lower than electronic
transducers in the wind tunnel tests. Generally, the differences between the manual and the
electronic devices are smaller on average and less variable in the wind tunnel tests than in the field
tests.
Wall Effects. Wall-effects adjustments to flow generally resulted in a decrease in calculated
velocity. The average percent difference in average velocity across all probe types and field test sites
for which complete wall effects data sets were obtained is -1.72% (s.d. = 0.23%, n = 5 sites). No
significant difference was found among probe types. A small difference was observed between the
percent difference in the average velocities measured in smooth stacks and rough stacks, but these
differences were not statistically significant at the 95% confidence level. The percent decrease in
average velocity between unadjusted and wall-effects adjusted traverses becomes smaller as the
number of points in the Method 1 traverse increases.
CALIBRATIONS
One- and Two-Dimensional Probes. Comparison of the NCSU pre-and post-test results showed that
all the post-test Cp values derived at 60 ft/sec were within ±1.5% of the pre-test values (the values
used to calculate flow in the field tests). The largest differences between pre- and post-test Cp values
occurred at 30 ft/sec. For all probes, the post-test 30 ft/sec Cp values were higher than the pre-test
coefficients, ranging from 0.4% higher for one Prandtl probe to 5.1% higher for one Type S probe.
Pre- and post-test Cp values obtained at 60 and 90 ft/sec were generally consistent. Even with the
comparatively large changes observed at 30 ft/sec, Cp values averaged over all three velocities
XI
-------
changed by less than 1.5% between pre- and post-test calibrations. Although the Cp values measured
at NIST were generally higher than the corresponding NCSU post-test values, the NIST post-test Cp
values were within ±2.2% of the NCSU values.
DAT Probes. For three of the four DAT probes, differences in velocities calculated between pre-
and post-test NCSU calibrations were less than 2% in the -20° to +20° pitch angle range. In the
pitch angle range of-10° and +10°, which is comparable to the range of pitch angles measured at
the three utility stacks during this field study, the differences in velocity were less than 1%. For
reasons that are unclear, the velocity differences ranged from 3.35% to 4.18% within the -10° to
+10° pitch angle range for the fourth DAT probe. A comparison of calculated velocities between
the NIST and post-test NCSU calibrations shows that, within the -10 ° to +10 ° pitch angle range, the
percent difference in calculated velocities over the three nominal wind tunnel velocity settings (i.e.,
30, 60, and 90 ft/sec) were 1.7% or less for each probe. In the -30° to +30° pitch angle range, the
average percent difference of calculated velocity between the NIST and post-test NCSU calibrations
was less than 3.8% for each probe.
Spherical Probes. For the four original spherical probes, the results of the four sets of calibrations
at NCSU showed that the calibration values varied by up to 5% over all pitch angles. For the second
set of probes, the change in calibration was approximately 4% to 5% within the -10° to +10° pitch
angle range. A comparison of the calculated velocities for the NIST and NCSU post-test spherical
probe calibrations shows that, within the -30° to +30° pitch angle range, the calculated NCSU
velocity for one probe was low relative to NIST (-1.2%). For all other probes, the NCSU calculated
velocities were 1.9% to 6.4% higher than those obtained by NIST.
Xll
-------
TABLE OF CONTENTS
Section Page
Executive Summary iii
Overview iv
1 Background and Goals of Study 1-1
1.1 Background 1-1
1.2 Goals of the Field Study 1-1
1.3 Approach 1-4
1.4 Key Characteristics of Field Test Sites 1-4
1.5 Document Organization 1-7
2 Data Screening Procedures, Probe Calibrations, and
Engineering Method Error Analysis 2-1
2.1 Data Screening Procedures 2-1
2.1.1 Data Excluded Due to Documented Equipment Problems 2-1
2.1.2 Statistical Evaluation of the Data 2-2
2.1.3 Evaluation of Outlier Probes 2-4
2.1.4 Process Stability Analysis 2-4
2.2 Verification of Probe Calibrations 2-8
2.2.1 Calibration Comparison: Pre-TestNCSU to Post-Test NCSU 2-9
2.2.2 Calibration Comparison: NIST to NCSU 2-11
2.3 Engineering Method Error Analysis 2-14
3 Analysis of Matrix A and Matrix B 3-1
3.1 Comparison Across Probes and Methods 3-1
3.1.1 Relationships Among Probes and Methods 3-1
3.1.2 Rank Order Analysis 3-10
3.1.3 Analysis of Variance of Matrix A Data 3-12
3.1.4 Analysis of Central Tendency 3-14
3.2 Within-Method Analysis: Analysis of Variance of Matrix B Data 3-17
3.2.1 ANOVA on Probe Copies 3-18
3.2.2 ANOVA on Test Teams 3-19
3.3 Analysis of Method Variability 3-20
3.4 Summary of Central Tendency and Variability Analyses 3-20
3.5 Uncertainty in Engineering Methods and Probe Measurements 3-23
4 Supplemental Data Analysis 4-1
4.1 Refinement of the Central Tendency Analysis 4-1
4.2 Comparison to Engineering Baseline 4-4
4.3 Sensitivity Analysis on Lake Hubbard Spherical Probe Data 4-6
4.4 Comparison of Manual and Electronic Pressure Measuring Devices 4-7
5 Wall Effects Study 5-1
5.1 Wall Effects Data Collection Procedure 5-1
5.2 Calculation of Replacement Velocity Values
Using Wall Effects Data 5-2
5.3 Summary of Wall Effects Data Obtained for the Study 5-2
5.4 Analysis of Wall Effects Data 5-5
5.4.1 General Findings 5-6
5.4.2 Point-to-Point Percent Change in Velocity 5-6
-------
TABLE OF CONTENTS (continued)
Section Page
5.4.3 Findings With Respect to Stack Material 5-9
5.4.4 Site-to-Site Comparisons 5-11
5.4.5 Findings Relating to the Number of Points in the Original Traverse 5-11
5.4.6 Probe Type Comparisons 5-12
5.4.7 Effect of Stack Gas Velocity 5-13
5.4.8 Findings on Practical Aspects of Wall Effects Testing 5-14
5.4.9 Wall Effects Calculations Using Minimum Number of Traverse Points ... 5-14
5.4.10 Maximum and Minimum Percent Differences 5-14
6 Findings 6-1
6.1 DAT Probe 6-1
6.2 Prandtl Probe 6-1
6.3 Spherical Probe 6-2
6.4 Autoprobes Yaw-Nulled 6-2
6.5 Autoprobes Straight-Up 6-3
6.6 Type S Probe Yaw-Nulled 6-3
6.7 Type S Probe Straight-Up 6-3
6.8 French Probe 6-4
6.9 Modified Kiel Probe 6-4
6.10 Comparison to Engineering Baseline 6-4
6.11 Comparison of Manual and Electronic Pressure Measuring Devices 6-5
6.12 Wall Effects 6-5
6.13 Calibrations 6-5
6.13.1 One- and Two-Dimensional Probes 6-5
6.13.2 Three-Dimensional Probes 6-5
7 Recommended Equipment and Procedures for Measuring Volumetric Flow 7-1
7.1 Background 7-1
7.2 Equipment Revisions 7-1
7.2.1 Probes and Probe Components 7-1
7.2.2 Probe Supports and Stabilization Devices 7-5
7.2.3 Yaw Angle-measuring Devices 7-6
7.2.4 Pressure-measuring Devices 7-6
7.2.5 Wind Tunnel Cross-sectional Area Requirements 7-7
7.3 Procedural Revisions 7-7
7.3.1 Wind Tunnel Procedures 7-7
7.3.2 Field Test Procedures 7-9
Appendix A: Stack Diagrams and Plots of Yaw and Pitch Angles
Appendix B: Probe Diagrams
Appendix C-l: North Carolina State University Pre- and Post-Test Probe Calibrations
Appendix C-2: National Institute of Standards and Technology Probe Calibrations
Appendix C-3: Supporting Data for First-Order Error Analysis of Engineering Methods
Appendix D: Rank Order Tables
Appendix E: Analysis of Variance
Appendix F: Confidence Interval Plots
XIV
-------
TABLE OF CONTENTS (continued)
Section Page
Appendix G: Dispersion Analysis
Appendix H: Initial Input Parameters and Results for Uncertainty Analysis
Appendix 1-1: Wall Effects Adjustment Factors Used in Round 2 Data Analysis
Appendix 1-2: Comparison of Manual and Electronic Pressure Measuring Devices
Appendix J: Wall Effects Data
XV
-------
LIST OF FIGURES
Figure Page
2-1 DeCordova: Volumetric flow by probe/method 2-2
2-2 Lake Hubbard: Volumetric flow by probe/method 2-3
2-3 Homer City: Volumetric flow by probe/method 2-3
2-4 Process stability at DeCordova, illustrated by volumetric flow (determined by
the baseline Autoprobe, 16-point, straight-up) and unit operating load 2-5
2-5 Process stability at Lake Hubbard (high load), illustrated by volumetric flow
(determined by the baseline Autoprobe, 16-point, straight-up) and
unit operating load 2-6
2-6 Process stability at Homer City, illustrated by volumetric flow (determined by
the baseline Autoprobe, 16-point, straight-up) and unit operating load 2-6
2-7 Process stability at DeCordova, illustrated by volumetric flow (determined
by the baseline Autoprobe, 16-point, straight-up) and O2 concentration
at the economizer outlet 2-7
2-8 Process stability at Lake Hubbard (high load), illustrated by volumetric flow
determined by the baseline Autoprobe, 16-point, straight-up) and O2
concentration at the economizer outlet 2-7
2-9 Process stability at Homer City, illustrated by volumetric flow (determined by
the baseline Autoprobe, 16-point, straight-up) and O2 concentration at
the economizer outlet 2-8
3-1 Flow percent difference from baseline Autoprobes 16-point,
straight-up at DeCordova (scale 1) 3-7
3-2 Flow percent difference from baseline Autoprobes 16-point,
straight-up at Lake Hubbard (scale 1) 3-7
3-3 Flow percent difference from baseline Autoprobes 16-point,
straight-up at Homer City (scale 1) 3-8
3-4 Flow percent difference from baseline Autoprobes 16-point,
straight-up at DeCordova (scale 2) 3-8
3-5 Flow percent difference from baseline Autoprobes 16-point,
straight-up at Lake Hubbard (scale 2) 3-9
3-6 Flow percent difference from baseline Autoprobes 16-point,
straight-up at Homer City (scale 2) 3-9
3-7 Summary of central tendency and variability results; All matrix A runs, all sites.... 3-22
5-1 Point-to-point percent change in velocity across all probes and ports
at DeCordova 5-7
5-2 Point-to-point percent change in velocity across all probes and ports
at Lake Hubbard 5-7
5-3 Point-to-point percent change in velocity across ports at Picway 5-8
5-4 Point-to-point percent change in velocity across ports at Mitchell 5-8
5-5 Point-to-point percent change in velocity across ports at Conesville Unit 1/2 Stack .. 5-9
5-6 Average and 95% confidence intervals of percent difference in velocity,
by stack material, based on equal weight of each test 5-11
5-7 Percent change in velocity due to wall effects vs. average velocity in the
original traverse 5-13
7-1 Error in test method produced by error in angle measurements 7-3
-------
LIST OF TABLES
Table Page
1-1 Cross-Site Test Matrix Summary: Number of Runs Performed by
Each Probe or Method 1-5
1-2 Key Physical Characteristics of Field Test Sites 1-6
1-3 Flow Characteristics of Field Test Sites 1-6
2-1 One- and Two-Dimensional Probe Calibration Coefficients Derived
by NCSU at 60 ft/sec Nominal Wind Tunnel Velocity 2-10
2-2 One- and Two-Dimensional Probe Calibration Coefficients Derived
at 60 ft/sec Nominal Wind Tunnel Velocity 2-12
2-3 Comparisons of Calculated Velocities Using the NIST and NCSU
Post-test Calibrations of DAT Probes 2-13
2-4 Comparisons of Calculated Velocities Using the NIST and NCSU
Post-test Calibrations of Spherical Probes 2-14
2-5 Pre- and Post-test Error Estimates of Engineering Methods 2-16
3-1 Overall Range in Flow Measurements 3-2
3-2 Range in Flow Measurements for In-stack Methods 3-2
3-3 Range in Flow Measurements for Engineering Methods 3-3
3-4 In-stack and Engineering Methods Comparison 3-3
3-5 Average Difference of Manual Type S Probe Straight-up from Engineering
Methods Matrix A and B Matrix Runs 3-4
3-6 Summary of Volumetric Flow Results—DeCordova 3-5
3-7 Summary of Volumetric Flow Results—Lake Hubbard 3-6
3-8 Summary of Volumetric Flow Results—Homer City 3-6
3-9 DAT, Autoprobe Yaw-Nulled, and MMBtu Performance: Percent Difference
from Baseline 3-10
3-10 Aggregate Rank Order of Volumetric Flow at all Sites Across all Methods
for Matrix A (Runs 1-8) 3-11
3-11 Duncan's Multiple Range Test on Probe Classes (Matrix A, a=0.05) 3-14
3-12 Central Tendency Analysis on Matrix A Runs at Each Site 3-16
3-13 Results of ANOVA and Multiple Comparison Tests to Detect Differences
Among Copies of the Same Probe Type in the Matrix B Data Set 3-18
3-14 Results of ANOVA and Multiple Comparison Tests to Detect Differences
Among Test Teams in the Matrix B Data Set 3-19
3-15 Method Variability Analysis for Matrix A Runs 3-21
3-16 Summary of Central Tendency and Variability Analyses: Probe Achievement
Levels 3-23
3-17 Monte Carlo Error Analysis Results on Volumetric Flow 3-25
4-1 Summary Statistics for Round 2 Central Tendency Analysis (Matrix A) 4-3
4-2 Summary Statistics for Round 2 Analysis Using MMBtu Method
as the Baseline (Matrix A) 4-5
4-3 Sensitivity Analysis on Round 1 Central Tendency Analysis for Matrix A
Runs—Lake Hubbard (High Load) (Based on Table 3-12) 4-7
4-4 Sensitivity Analysis on Method Variability Analysis for Matrix A Runs
(Based on Table 3-15) 4-8
XVI1
-------
LIST OF TABLES (continued)
Table Page
4-5 Sensitivity Analysis on Volumetric Flow Summary—Lake Hubbard
(Based on Table 3-7) 4-9
4-6 Sensitivity Analysis on Duncan's Multiple Range Test—Lake Hubbard
(High Load) (Based on Table 3-11) 4-10
4-7 Sensitivity Analysis on Overall Range of Flow Measurements—Lake
Hubbard High Load) (Based on Table 3-1) 4-10
4-8 Sensitivity Analysis on In-stack and Engineering Methods Comparison—
Lake Hubbard (High Load) (Based on Table 3-4) 4-11
5-1 Key Characteristics of Sites for Additional Wall Effects Tests 5-3
5-2 Summary of Percent Difference Between Original and
Wall Effects—Adjusted Average Velocity 5-5
5-3 Summary Statistics on Percent Difference in Velocity for Steel Stacks
and for Brick and Mortar Stacks Based on Unaggregated Data andl 1
Data Aggregated by Site 5-10
5-4 Summary of Unadjusted and Wall Effects-Adjusted Velocities for
16- and 12-Point Traverses 5-12
5-5 Comparison of Percent Differences Between Original and Wall Effects-
Adjusted Average Velocities, Using All Data Points and Using Only
Two Data Points per Port 5-15
7-1 Proposed Yaw Angle Tolerances 7-2
7-2 Proposed Pitch Angle Tolerances 7-2
7-3 Proposed Velocity Calibration Tolerances 7-2
7-4 Horizontal Straightness Declination Limits 7-10
-------
SECTION 1
BACKGROUND AND GOALS OF STUDY
1.1 BACKGROUND
This report describes an experimental program sponsored by the U.S. Environmental Protection
Agency (EPA) to evaluate potential improvements to the Agency's current reference method for
measuring volumetric flow (Method 2,40 CFR Part 60, Appendix B). Method 2 [Determination of
Stack Gas Velocity and Volumetric Flow Rate (Type S Pitot Tube)] specifies measurements to
determine volumetric flow, but does not prescribe specific methods to account for yaw or pitch
angles of flow when the flow in the stack is not axial. Method 2 also allows the use of only two
probe types, the Type S and the Prandtl.
During the summer of 1997, EPA conducted week-long field tests at three electric utility sites to
evaluate potential improvements to Method 2. The candidate improvements grew out of a technical
review of Method 2 and draft Method 2F (an unpromulgated, mid-1993 proposal for using three-
dimensional probes in situations where flow velocity has significant yaw and pitch components),
extensive preliminary wind tunnel and field testing, and comments provided by industry on the
performance of currently available technology.
Three test sites were chosen to provide a range of flow conditions under which the tested equipment
and procedures could be rigorously evaluated. One site had near-axial flow, another had flow with
moderate yaw and pitch angles, and the third had flow with a significant yaw component. Two sites
were gas-fired and the third was coal-fired, which provided an opportunity to compare in-stack
measurements of gas flow with calculations of volumetric flow using engineering methods for more
than one type of fuel. "Wall effects" tests were conducted at the three primary sites and six
additional stacks to measure the impact of velocity decay near the stack wall on volumetric flow.
The characteristics of the three field test sites are summarized in Section 1.4. Stack diagrams
showing port and traverse point locations are presented in Appendix A. A full description of each
site and the specific tests performed appears in the separate site data reports for each site1. Tests
were run at six additional stacks to measure the impact, if any, of the "wall effect" on flow
measurement accuracy.
1.2 GOALS OF THE FIELD STUDY
The field tests had four primary goals. The first was to identify changes to Method 2 procedures that
could improve the accuracy of the method under various flue gas flow conditions. The second was
1 The Cadmus Group, Inc. 1998," EPA Flow Reference Method Testing and Analysis: Data Report, Texas Utilities, DeCordova
Steam Electric Station, Volume I: Test Description and Appendix A (Data Distribution Package)," EPA/430-R-98-015a.
The Cadmus Group, Inc. 1998, " EPA Flow Reference Method Testing and Analysis: Data Report, Texas Utilities, Lake
Hubbard Steam Electric Station, Volume I: Test Description and Appendix A (Data Distribution Package)," EPA/430-R-98-
017a.
The Cadmus Group, Inc. 1998," EPA Flow Reference Method Testing and Analysis: Data Report, Pennsylvania Electric Co.,
G.P.U. Genco Homer City Station: Unit 1, Volume I: Test Description and Appendix A (Data Distribution Package),"
EPA/430-R-98-018a.
1-1
-------
to ensure that each contemplated change would result in measurements that have acceptably low
variability. The third goal was to ensure that each change would be practicable for use in the field
and able to withstand the conditions typically encountered in electric utility stacks. The fourth was
to ensure that any changes to the methods would not result in harm to the environment because of
systematically low measurements of volumetric flow. In addition to these four primary goals, two
collateral goals were identified. One was to seek improvements to Method 2 procedures that might
reduce the disparity that industry had reported to occur under certain circumstances between the heat
rate derived from in-stack measurements of flow and heat rate calculated from combustion
parameters. The second was to collect data that would quantify the "wall effects," that is, the decline
in flow velocity near the stack wall.
The field study addressed the four primary goals by evaluating the performance of nine
probe/procedure combinations considered for inclusion in a revision to Method 2. The nine tested
probe/procedure combinations were the manual Type S probe straight-up (the current Method 2
procedure), manual Type S probe yaw-nulled, automated Type S probe straight-up, automated
Type S probe yaw-nulled, DAT, spherical, modified Kiel, French, and Prandtl. Probe types similar
to those used in the field tests were initially qualified for inclusion in the study through testing
performed during 1996 at the Merrill Subsonic Wind Tunnel at North Carolina State University
(NCSU) in Raleigh2. All of the probes used in the field tests were provided to the test teams by EPA,
except for the DAT probes, which were supplied by the test teams. Diagrams of each probe are
presented in Appendix B. The probe/procedure combinations are described below.
The manual Type S probe has been used extensively to measure flow in accordance with the testing
requirements of 40 CFR Part 75. When used in accordance with the procedures currently prescribed
in Method 2, the probe is referred to as the "straight-up" probe; that is, it is positioned so that an
imaginary line connecting the impact port and the static port is parallel to the longitudinal axis of the
stack. In this configuration, it measures total velocity. During the test program, the manual Type S
probe was also operated in the "yaw-null" mode that included determining the yaw component of
the flow.
An automated version of the Type S probe, manufactured by United Sciences Testing, Inc.—the
USTI Autoprobe Type S probe—was also evaluated in both the straight-up and yaw-nulled modes.
Two configurations of the Autoprobes were operated in this field study, a single Autoprobe was
moved from port to port like the other tested pitots, and a four-probe configuration (referred to as
the "baseline" Autoprobes) was mounted in four separate ports and collected data from these
locations for the duration of the field tests.
3-D probes provide a measure of both yaw and pitch angles of gas flow using five ports on the
sensing head of the probe. For this field study, two types were tested: the DAT and the spherical
probes. The DAT probe is prescribed in draft Method 2F. The spherical probe was recommended
for consideration by the Electric Power Research Institute (EPRI). It is not commonly available.
EPA had the spherical probes used for this study specially fabricated in stainless steel. Two sets of
The Cadmus Group, Inc. 1997, "Flow Reference Method Testing and Analysis: Wind Tunnel Experimental Results,"
EPA/430/R-97-013.
1-2
-------
four spherical probes, a total of eight probes, were ultimately fabricated after the welds on the first
set sustained thermal damage during the first field test.
The modified Kiel probe is a standard Kiel probe with a wake port and two Fechheimer ports added.
The modified Kiel probe was suggested for use by EPA and is not available commercially. EPA
therefore had copies of the probe specially fabricated for this study. A prototype modified Kiel probe
was tested in the NCSU wind tunnel to determine whether this probe was sufficiently promising to
warrant further consideration as an acceptable method for measuring flow. Results obtained in the
wind tunnel experiments were used in finalizing the probe's design for use in the field testing.
Recommended for consideration by EPRI, the French probe is a cylindrical probe with a solid
stainless steel probe head, an impact port with a 15 ° chamfer on the probe head, and a wake port
located at a 180° offset from the impact port. This probe head design does not meet current
specifications in Method 2. The probe, designed by Jennifer French of Southern Company Services,
is not commercially available. EPA therefore had copies of the probe specially fabricated for this
study.
The Prandtl probe is the accepted standard that is used both to calibrate flow rates in wind tunnels
and to establish calibration coefficients for other types of pilot tubes. Also called the "standard
pitot," the Prandtl probe used in this project was a 0.25-in. outside diameter stainless steel tube with
a 90° bend located approximately 3.5 in. from the hemispherical probe tip. Four, 0.035-in. inside
diameter static pressure ports are located 1 in. from the probe tip and positioned 90° apart around
the tube.
In conjunction with testing the probes, new field calibration and operational procedures were
performed and evaluated as potential method improvements. Digital inclinometers were also tested
to evaluate these angle measuring devices for performing yaw-nulling, rotational position calibration,
and horizontal straightness checks. A laser device was used to measure stack diameter, and the
results were compared to values obtained by calculating the diameter from measurements of the
circumference of the stack.
This study was designed to test in-stack methods (i.e., the probes currently allowed or being
considered for inclusion in new draft methods). However, to address industry's concern over
instances of disparity between heat rates based on conventional industry calculational methods and
heat rates based on continuous emission monitoring system (CEMS) measurements, five engineering
methods, not under consideration for inclusion in new methods, were included in the study. These
five engineering methods were used to evaluate the extent of the disparity between the in-stack and
engineering approaches to calculating heat rate. To reduce the sources of variability in these
engineering methods, great effort was taken to ensure process stability, collect a vast array of
combustion and process parameter data, perform frequent fuel sampling, and obtain multiple,
independent fuel analyses. Several specific measures were taken to ensure that the engineering
methods were implemented as rigorously as possible. First, non-routine plant operating data were
required at each site. Examples of these non-routine data include continuous automated natural gas
sampling with analysis by an on-line gas chromatograph, described in the site data reports for the two
gas-fired sites. At the coal-fired site, separate coal samples were taken corresponding to each stack
run, and laboratory analysis was performed on each sample. Second, unit load was held steady
1-3
-------
during the entire week of testing at each site to accommodate the different response times between
the engineering methods and the in-stack measurements. Finally, a specialist from Lehigh
University's Energy Research Center (ERC) assembled the necessary data and performed the
engineering volumetric flow calculations. Copies of the combustion and process parameter data
were also provided to field test observers representing EPRI, so that they could offer an independent
spot check of the engineering data and calculations.
1.3 APPROACH
A collaborative field testing protocol was used under which multiple stack-testing teams performed
simultaneous measurements using Method 2 and pre-selected alternatives to Method 2. The
collaborative (or multi-team) approach ensured that the data collected on each tested alternative are
representative of and support an assessment of the extent and sources of variability in the
measurements. Concurrent with the test team measurements, in-stack measurements were made
using the Autoprobe and using a cross-correlation infrared monitor. The cross-correlation monitor
was being evaluated as an adjunct to this study. Simultaneously with the in-stack measurements,
plant parameter data were collected to enable volumetric flow to be estimated using the five
engineering methods.
A three-part experimental design was used in the field tests:
1. Matrix A—In this interprobe comparison, simultaneous measurements from different
types of probes were compared to examine the relative magnitudes of volumetric flow
measurements determined by different probe types.
2. Matrix B—In thi s intraprobe comparison, multiple copies of the same types of probes were
compared. In conjunction with Matrix A, Matrix B was designed to provide more
extensive data on the variability in each probe's measurements.
3. Matrix C—The wall-effects study prescribed the collection of data in 18 1 -in. increments
from the stack wall. This study was conducted to evaluate the extent of decay in flue gas
velocity near the stack wall.
Table 1-1 provides an overview of Matrices A, B, and C, showing the methods tested and the
number of runs performed at each site. A run consisted of each of the four test teams concurrently
performing a full 16-point stack traverse. Simultaneous traverses were performed by the four
baseline Autoprobes that were installed in separate ports. A full description of the experimental
design used at each test site is found in each site data report.
1.4 KEY CHARACTERISTICS OF FIELD TEST SITES
Characteristics of the three primary field sites are presented in Table 1-2. Test dates and average
velocity, yaw angles, and pitch angles, measured during the tests are presented in Table 1-3.
DeCordova has the highest nameplate capacity, followed by Homer City, and then Lake Hubbard.
Operating loads during the test were generally maintained near each unit's maximum capacity. Four
low-load runs were also conducted at the Lake Hubbard site.
1-4
-------
Table 1-1. Cross-Site" Test Matrix Summary: Number of Runs Performed by Each
Probe or Method
Method
DAT
Type S Straight-up
Type S Yaw-nulled
Modified Kiel
Spherical Ab
Spherical Bb
Prandtl
French
Manual Autoprobe 16-point Straight-up
Manual Autoprobe 16-point Yaw-nulled
Baseline Autoprobes 12-point Straight-up
Baseline Autoprobes 12-point Yaw-nulled
Baseline Autoprobes 16-point Straight-up
Baseline Autoprobes 16-point Yaw-nulled
Baseline Autoprobes 20-point Straight-up
Baseline Autoprobes 20-point Yaw-nulled
Baseline Autoprobes 48-point Straight-up
Baseline Autoprobes 48-point Yaw-nulled
Infrared Monitor Ac
Infrared Monitor Bc
O2 F-Factor
CO2 F-Factor
MMBtu
ASMEPTC4.1
BTCE
Installed CEMS
Matrix A
DC
8
4
4
4
3
—
4
4
4
4
—
—
8
8
—
—
8
8
8
—
8
8
8
8
8
—
LH
High
Load
9
5
5
5
5
—
4
4
4
4
—
—
9
9
—
—
8
8
1
—
9
9
9
9
9
—
Low
Load
4
4
4
4
4
—
—
—
—
—
—
—
4
4
—
—
4
4
4
—
4
4
4
4
4
—
HC
8
4
4
4
4
4
—
4
4
4
8
8
8
8
0
0
0
0
8
8
8
8
8
8
8
8
Matrix B
DC
4
4
4
4
4
—
4
4
—
—
—
—
23
14
—
—
23
14
8
—
24
24
24
24
24
—
LH
4
4
4
—
4
—
—
—
—
—
—
—
12
12
—
—
12
12
12
—
12
12
12
12
12
—
HC
4
4
4
4
4
4
—
4
—
—
9
9
24
20
15
11
—
—
24
24
24
24
24
24
24
24
Matrix C
DC
1
1
1
1
1
—
1
1
0
0
—
—
6
4
—
—
5
3
—
—
—
—
—
—
—
—
LH
1
—
—
—
—
—
—
—
—
—
—
—
1
1
—
—
1
1
—
—
—
—
—
—
—
—
HC
1
1
1
1
1
1
—
1
—
—
2
2
6
5
4
3
—
—
—
—
—
—
—
—
—
—
DC = DeCordova, LH = Lake Hubbard, HC = Homer City.
Four additional spherical probes were tested at Homer City; the original set of spherical probes was designated Spherical
"A," and the additional set was designated Spherical "B."
Two (redundant) infrared monitors (designated as A and B) were tested at Homer City.
1-5
-------
Table 1-2. Key Physical Characteristics of Field Test Sites
Characteristic
Location
Fuel
Boiler Type
Nameplate Capacity
Load Range During Tests
Inside Stack Diameter
Stack Height
Test Port Distance from
Downstream Disturbance
Test Port Distance from
Upstream Disturbance
DeCordova
40 mi. S.W. of
Fort Worth, TX
Natural gas
Babcock & Wilcox
5,445,000 Ib/hr
opposed-fired
799 MWe
7ft/l \yf\X7** 7KA \yf\l/A
23.08 ft
209ft
28.1ft(1.2diam.)
71.8 ft (3.1 diam.)
Lake Hubbard
20 mi. E. of
Dallas, TX
Natural gas
Babcock & Wilcox
3,825,000 Ib/hr
opposed-fired
531 MWe
508MWe-510MWe
(high-load runs)
251 MWe
(low-load runs)
22.5ft
199ft
53.5 ft (2.4 diam.)
46.0 ft (2.0 diam.)
Homer City
Homer City, PA
Bituminous coal
Foster Wheeler
4,620,000 Ib/hr
opposed-fired
660 MWe
«Q4 MWp— 6S5 MWp
24.0ft
800ft
420 ft (17.5 diam.)
199 ft (8.3 diam.)
Table 1-3. Flow Characteristics of Field Test Sites
Characteristic"
Test Dates
Velocity (ft/sec) (mean ± s.d.)
Average Absolute Yaw Angle (°) (mean ± s.d.)
Average Absolute Pitch Angle (°) (mean ± s.d.)
Average Yaw Angle (°) (mean ± s.d.)
Average Pitch Angle (°) (mean ± s.d.)
DeCordova
6/3/97-6/8/97
94.05 ± 6.24
2.34 ± 2.00
3.42 ±2.49
-0.54 ± 3.03
2.71 ±3.24
Lake Hubbard
7/8/97-7/12/97
65.36 ±17.78 (high load)
39.13 ±10.65 (low load)
6.84 ± 6.01 (high load)
7.28 ± 6.40 (low load)
4.01 ±3. 60 (high load)
4.53 ± 4.47 (low load)
-0.21 ±9. 10 (high load)
-0.82 ± 9.67 (low load)
3. 19 ±4.34 (high load)
2.59 ± 5.83 (low load)
Homer City
8/19/97-8/24/97
77.89 ±4.35
13.93 ±5. 14
2.24 ±1.87
13.93 ±5.14
1.17 ±2.67
Equations for calculating these averages are as follows:
Average Absolute Yaw Angle =
Average Yaw Angle =
where:
Y, = individual measured yaw angle,
P, = individual measured pitch angle,
n = number of measurements, and
s.d. = standard deviation.
-^ 1 s.a., average A.DSOIUIC
n
rucn /\ngie —
n
I>
'=1 + r A
n
±s.d.
1-6
-------
Stack diameter was approximately the same at all three sites (22-24 ft). Stack height was
approximately 200 ft at the two gas-fired sites and 800 ft at the coal-fired site. Average flue gas
velocity (as an average measured by all probe types) determined during the high-load runs was
approximately 95 ft/sec at DeCordova, 65 ft/sec at Lake Hubbard, and 78 ft/sec at Homer City.
During the low-load runs at Lake Hubbard, flow velocity averaged about 39 ft/sec.
At DeCordova, the average absolute yaw angle was 2.34° with a standard deviation of 2.00°. The
average absolute pitch angle was 3.42° with a standard deviation of 2.49°. Flow measurements
obtained from each of the ports were generally uniform. Because the test location was within 1.2
diameters of the stack exit, wind gusts over the top of the stack occasionally caused pulses in the
stack pressures.
The flow profile at Lake Hubbard was quite different from the nearly axial flow at DeCordova. Gas
flow at the Lake Hubbard stack was stratified, with significantly higher gas velocities at the back of
the stack where Ports B and C were located than those measured at the front of the stack (see the
relatively high standard deviations for the velocity values presented in Table 1-3). Note that the
relatively high standard deviations for the yaw angles at Lake Hubbard are due to the large variation
in yaw angles; the high standard deviation does not reflect measurement error. The flow may also
have exhibited a double swirl pattern, as indicated by the measured yaw angles in each port; the yaw
angles were generally positive in Ports A and B and generally negative in Ports C and D. The
average absolute yaw angles were 6.84° with a standard deviation of 6.01 °over the high-load runs
and 7.28° with a standard deviation of 6.40° over the low-load runs. The average absolute pitch
angles at Lake Hubbard were 4.01 ° with a standard deviation of 3.60° over the high-load runs and
4.53 ° with a standard deviation of 4.47° over the low-load runs.
The test ports on the Homer City stack were installed at a location that satisfies Method 1 criteria
for 12-point traverses (8 diameters downstream and 2 diameters upstream from the nearest flow
disturbance). The average absolute yaw angle was 13.93° with a standard deviation of 5.14 °. The
average absolute pitch angle was 2.24° with a standard deviation of 1.87°.
Box-and-whisker plots included in Appendix A illustrate the yaw and pitch angles at each site.
1.5 DOCUMENT ORGANIZATION
This section (Section 1) describes the background and goals of EPA's field flow test program. The
section is supported by Appendix A, which illustrates the stack cross-sections and flow
characteristics at the DeCordova, Lake Hubbard, and Homer City test sites, and Appendix B, which
presents diagrams of the probes tested in this study.
Section 2 describes the procedures used to develop a final data set for analysis, to compare how
changes in the calibration coefficients derived before and after the field tests affect volumetric flow
results, and to evaluate the error in flow estimates derived from the five engineering methods.
Appendix C provides the supporting information for the calibration coefficient comparison
(Appendices Cl and C2) and the error analysis of the five engineering methods (Appendix C3).
Section 3 presents the results of the first round of the data analyses (performed before this document
was peer reviewed). The section has several major subsections, including comparisons of results
across methods (3.1); comparisons of results within methods (3.2); an analysis of method variability
(3.3); a summary of central tendency and variability analysis (3.4); and a comparison of the extent
of uncertainty inherent in in-stack versus engineering methods (3.5). Five appendices provide
1-7
-------
supporting information and data for Section 3: Appendix D, Rank Order Tables; Appendix E:
Analysis of Variance; Appendix F, Confidence Interval Plots; Appendix G, Dispersion Analysis; and
Appendix H, Input Parameters and Results for Uncertainty Analysis.
Section 4 presents the results of the supplemental data analyses, which were conducted based on
recommendations made to EPA by a panel of peer reviewers. The results of a refined central
tendency analysis are presented in Section 4.1; Section 4.2 presents a comparison of probe
measurements to an engineering baseline; and Section 4.3 presents the results of a sensitivity analysis
to determine the effect on the conclusions of excluding certain data points that appeared implausible
to the peer reviewers. Section 4.4 presents the results of a comparison of manual and electronic
pressure measuring devices, using data from four sources: (1) wind tunnel tests conducted at the
National Institute of Standards and Technology (NIST); (2) wind tunnel tests conducted by the Fossil
Energy Research Corporation (FERCo) at their swirl tunnel; (3) field test data collected by Radian
Corporation at an electric utility plant in Pennsylvania; and field test data collected by FERCo at the
Coal Creek Power Plant, located in Washburn, North Dakota and the Columbia Power Plant, located
in Portage, Wisconsin. Supporting data for the analyses in this section are provided in Appendix I.
Section 5 presents a detailed discussion of wall effects. Supporting information is provided in
Appendix J.
Section 6 presents a discussion of the major findings of this study. Findings are presented by probe
type. Additional findings are presented on probe calibration, wall effects, and other issues.
Section 7 presents recommended revisions to equipment and procedures used to determine
volumetric flow. The recommendations are based on the major findings discussed in Section 6 and
field test observations.
1-8
-------
SECTION 2
DATA SCREENING PROCEDURES, PROBE CALIBRATIONS, AND
ENGINEERING METHOD ERROR ANALYSIS
This section describes preparatory activities performed on the data before the first round of analysis.
Section 2.1 presents the systematic data screening procedures used to identify outlier volumetric flow
measurements. Section 2.2 discusses the procedures used to verify probe calibrations. Section 2.3
describes the error analysis performed on the engineering methods. As a result of peer review of the
procedures described in this section and the analyses presented in Section 3, additional screening
activities were conducted as part of supplemental sensitivity analyses of the field test data. These
additional screening activities and the resulting analyses are presented in Section 4.
2.1 DATA SCREENING PROCEDURES
A three-step, systematic procedure was implemented to identify individual volumetric flow values
or specific methods that should not be included in data analyses due to data quality flaws or to
problems in the operation of the method. The first step was visual inspection of plots of volumetric
flow data by run; the second was the application of a classical statistical tool to identify possible
outliers; and the third was visual inspection of plots of volumetric flow, O2 concentration at the
economizer outlet, and unit operating load (MWe) for each run at the three sites.
In the first step of the data screening process, plots of volumetric flow versus run were examined
visually to identify measurements that appeared to differ substantially from other values in the same
or proximal runs. Raw data records and supporting information associated with all values thus
identified were then reviewed. Any indication of errors or other data quality problems, including
data recording errors, transcription errors, errors in calibration data, calculational errors, equipment
malfunctions, or procedural problems was noted. Correctable errors identified through review of
data records and supporting information were corrected. Values that were flawed due to documented
procedural or equipment problems were excluded from analysis.
After this initial screening was completed, a statistical tool—the interquartile range, or IQR—was
used to identify additional suspicious values. The IQR is the difference between the 25th percentile
value and the 75th percentile value of all measurements taken for a specific method. The IQR is a
standard metric for representing the range of values within which the majority of the measurements
occur. Values falling outside of 1.5 times the IQR were flagged, and the raw data and supporting
information were reviewed. The 1.5-times-the-IQR criterion was also used to examine individual
methods that substantially differed from other methods.
Finally, the stability of each unit's operation during the course of the experiment was evaluated by
plotting volumetric flow, O2 concentration at the economizer outlet, and unit operating load for each
run for each site. The resulting effect of unit instability on measured volumetric flow values was
examined.
2.1.1 Data Excluded Due to Documented Equipment Problems
Review of raw data records and supporting information for the values identified through initial visual
examination of data plots disclosed that two volumetric flow values were flawed and therefore
2-1
-------
inappropriate for analysis. The two values were (1) the value obtained by the modified Kiel probe
(MK-4) for Run 2 at Lake Hubbard and (2) the value obtained by one of the DAT probes for Run
11 at Homer City. The former was flawed because the probe was bent during the run. The bend was
eliminated before the only other use of the MK-4 probe at Lake Hubbard in Run 23. The latter was
flawed because moisture was discovered in some of the probe lines after the run. This problem was
not observed for any other runs involving the DAT probe at Homer City. These two volumetric flow
values, and all data used to generate these values, were excluded from any further analysis.
2.1.2 Statistical Evaluation of the Data
Figures 2-1 through 2-3 present box-and-whisker plots of the volumetric flow values (including
Matrix A and Matrix B values, but excluding the two flawed values identified in Section 2.1.1) for
each probe at the three field test sites. Values identified on the plots with an asterisk (*) fall outside
the "whiskers" representing the 1.5-IQR value range for a specific probe.
When applied to the probe and engineering method data, the 1.5-IQR criterion resulted in the
flagging of 4 values at DeCordova, 5 values at Lake Hubbard, and 18 values at Homer City. All 4
of the values at DeCordova and 11 of the 18 values at Homer City were ones that had not been
identified by visually examining data plots; all 5 of the values at Lake Hubbard were ones that had
been identified by visually examining data plots. The raw data and supporting information were
reviewed for any indication of errors, underlying technical or engineering problems, or other data
quality problems, as discussed above.
1,80O,OOO
« 1,70O,
1
70O,OOO
i ,600,000
3 1,500,000
o
1,400,000
"
Figure 2-1. DeCordova: Volumetric flow by probe/method (horizontal lines represent the
minimum and maximum 1.5-IQR values for all methods except infrared monitor).
2-2
-------
1
1
Ǥ i
o
JD
O
I
30O.OOO
,200,000 -.
,100,000
,000,000
900,000 -.
800,000 -
700,000
600,000
500,000
400,000
300,000 -1
Figure 2-2. Lake Hubbard: Volumetric flow by probe/method (horizontal lines represent
minimum and maximum 1.5-IQR values for all methods except infrared monitor).
2,300,000 -
2,200,000 -
x-v 2.1OO.OOO T
° 1,3OO,OOO -.
1 20O 000 J
1,1OO,000 J
•
T
D
I
S*?i^i5 B ^*§jR I I § ^ 5
|jj - « I I I T
. *
'''"'////////#/'"
Figure 2-3. Homer City: Volumetric flow by probe/method (horizontal lines represent minimum
and maximum 1.5-IQR values for all methods except infrared monitor).
2-3
-------
Examination of the raw data and supporting information showed no engineering or data-based
reasons for identifying the values as flawed. Therefore, because no plausible reason for excluding
the values from further analysis was evident, the values were retained in the analytical data set.
2.1.3 Evaluation of Outlier Probes
Examination of the probe-specific box-and-whisker plots shows that the infrared monitor values are
not consistent with the measurements resulting from the other probes and engineering methods (see
Figures 2-1 through 2-3). As a test of the difference between the infrared monitor values and the
other methods, the minimum and maximum 1.5-IQR values obtained across all other methods,
excluding the infrared monitor, were identified. The minimum and maximum 1.5-IQR values are
represented by the dashed horizontal lines in Figures 2-1 through 2-3. As seen in these figures, the
infrared monitor consistently produced values that were below the minimum or above the maximum
1.5-IQR values.
The infrared monitor values were consistently and substantially higher than values obtained by the
other methods at DeCordova (Figure 2-1). At Lake Hubbard, however, the infrared monitor
produced volumetric flow values that were always substantially lower than those by the other
methods (Figure 2-2). Investigation of the apparent problem indicated by the Lake Hubbard field
test data led to the installation and operation of two infrared monitors for the third field test at Homer
City. Nonetheless, as shown in Figure 2-3, the average values produced by the two infrared monitors
at Homer City were consistently and substantially higher than values obtained by the other methods.
In addition to these data quality problems, the infrared monitors produced less than complete data
for two of the field tests. The infrared monitor failed to produce values for 20 runs at the DeCordova
field test and for eight runs at the Lake Hubbard field test.
Taken together, the large variability of the infrared monitor values, the large and erratic differences
between these values and values produced by other methods, and the incomplete data sets suggest
unresolved technical problems with the application of this technology in measuring volumetric flow
under the conditions prevailing at the three test sites. To have included the infrared monitor values
in the overall analysis of field test data would have produced skewed and unrepresentative results.
For these reasons, the data from the infrared monitors have been presented in the site data reports,
but are not included in the analyses for this report.
2.1.4 Process Stability Analysis
Maintaining process stability was an important consideration in attempting to limit the introduction
of confounding factors in the comparison of flue gas flow rates from different methods. Process
stability was particularly important for comparing in-stack and engineering methods because of the
likelihood that the two approaches would display different response times in the presence of transient
operating conditions. Thus, the field test data were screened to eliminate data collected during
unstable conditions. Unit operating load, O2 concentrations in the flue gas at the economizer outlet,
and volumetric flow measurements taken by the baseline Autoprobes (operated in the straight-up
mode) were used as indicators of process stability. The process was considered to be stable during
a run if load was within ±2.5% of the corresponding values measured during Run 1. (Run times
varied between approximately 45 and 75 minutes.)
2-4
-------
Figures 2-4 through 2-6 show the relationship between load values and volumetric flow
measurements taken by the baseline Autoprobe for each run at DeCordova, Lake Hubbard, and
Homer City, respectively. Figures 2-7 through 2-9 show the relationship between O2 concentration
at the economizer outlet and volumetric flow (also measured by the Autoprobe) at the three sites.
At DeCordova (Figures 2-4 and 2-7), no substantial change in the operation of the power plant
during the time of the experiment is indicated. The maximum absolute deviation from the first run's
flow and load values were 0.31% (Run 29) and 0.12% (Run 32), respectively. Plant operating
conditions therefore have no substantial impact on the interpretation of the analytical results or on
the data chosen for analysis.
CO
1
5
1800
1780
1760
1740
1720
1700
1680
1660
1640
1620
1600
1580
1560
1540
1520
1500
_
_
_
_
_ -
-
_
A + *. + + + + >A> + + + + + ^ + A + + + + + + +^>+^^<> + + >>+ +
-
-
_
I I I I I ! I I I I I I I ! I I I I I I I I I I I ! I I ! ! I ! I I I I II
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38
Run No.
•*" Volumetric Flow
-•-Load
900
880
860
840
820 5
800 S.
•o
,-„ «J
780 o
760
740
720
700
Figure 2-4. Process stability at DeCordova, illustrated by volumetric
flow (determined by the baseline Autoprobe, 16-point,
straight-up) and unit operating load.
The load and O2 values at Lake Hubbard during the "high-load" operation (Runs 1-20 and Run 26)
also show no substantial change in the operation of the power plant during the time of the experiment
(Figures 2-5 and 2-8, respectively). [Run 21, a wall effects run conducted at high load, was not
included in the process stability analysis because this run was analyzed separately as part of the wall
effects analysis (see Section 5)]. With Runs 21-25 excluded, the maximum deviations from the first
run's flow and load values were 2.0% (Run 9) and 0.47% (Run 14), respectively.
As indicated in the experimental design section, Runs 22 through 25 are low-load (~250 MWe)
replicates of Runs 1 through 4, which were performed at a high-load condition. Due to this change
in load conditions, Runs 22 through 25 are evaluated individually in all subsequent data analyses.
Isolating these values from the measurements taken at all other runs minimizes the influence of load
on the interpretation of the analytical results. In addition, Run 26 is a replicate of Run 1, taken after
the plant regained full-load conditions. However, testing was suspended after Run 26, resulting in
an unbalanced set of measurements relative to the experimental design. Therefore, for those analyses
that require an equal number of samples per probe (see Section 3 for a discussion of balanced
designs), Run 26 was dropped from consideration. For those analyses that are descriptive in nature
(e.g., calculation of means and variances), Run 26 was included in the analytical data set.
2-5
-------
8
O.
3
—
1 1
ii
U. h-
0)
3
1
1300 "fn
1280
1260
1240
1220
1200
1180
1160
1140
1120
1100
1080
1060
1040
1020
innn
-
-
-
04444444 — + 4. — 4 — 4 4 — 4- — 44444* — 4 41
""^
-
-
i ; : i ; ; i i i i i i i ! i i i i
580
560
540
520 ?
500 5.
480 §
460
440
420
400
1 3 5 7 9 11 13 15 17 19 21 23 25
2 4 6 8 10 12 14 16 18 20 22 24 26
Run No.
"•" Volumetric Flow
-•- Load
Figure 2-5. Process stability at Lake Hubbard (high load), illustrated by
volumetric flow (determined by the baseline Autoprobe, 16-
point, straight-up) and unit operating load.
S
0.
1
•p-
i!
II
p
E
.3
>
1600 7nn
1580
1560
1540
1520
1500
1480
1460
1440
1420
1400
_
_
-»-»-»-«^»-*-»^4~4^*~'~*~*^ T-*"*^*-*-^
- *-*-»-*-• ~* * * * * \ I ^~^~*~~*> T~*~*~*
_ \ \ /
- ^.». . m. MM v ' m f * ;
'"''' "' " if "" // :
1380 L II ~
1360 L '
r
1340 }- • "
1**9n [ i I I I , I ] I | I ! I | I I I | I I I I i ! | j I I I I I I ! | | I I t
680
660
640
620
600
580
560
540
520
500
1 3 5 7 9 11 13 15 17 19 21 23 25B 27 29 31 33 35 37
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38
Run No.
~*- Volumetric Flow
~*- Load
^^
i
I
Figure 2-6. Process stability at Homer City, illustrated by volumetric
flow (determined by the baseline Autoprobe, 16-point,
straight-up) and unit operating load.
2-6
-------
CO
u>
Q.
CD
J^
"P •
S 1
^ 5
O
1
5
-*
1800 'n
1780
1760
1740
1720
1700
1680
1660
1640
1620
1600
1580
1560
1540
1520
1500
_
_
-
-
_
- ^^^-^^r^^^^f^^^^^^^»^^^^^^^^^^^^^^^
• »"* ^•^-*-*-^-*-^-*-»-*-* • «4t»»««« •-»-» #-^
• •-.. •••—•••• ••••••••• "
•*••• ••••• • • •
_
I I I I I I I I I I I I I I I I I ! I I j I I ! I I I I I I I I I I I I I
3.5
3.0
2.5
2.0
1 5
1.0
0.5
0.0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38
Run NO.
~"~ Volumetric Flow
~"*~ 02, Economizer Outlet
g
^
O
3
E
i
&
CM*
o
Figure 2-7. Process stability at DeCordova, illustrated by volumetric
flow (determined by the baseline Autoprobe, 16-point,
straight-up) and O2 concentration at the economizer
outlet.
U)
£ 1280 L
gj 1240
•§ 1200
t5 45
| | 1160 -
§ 1 1120 -
S 1080
J 1040
> 1000
'*—»—»—*—*
4.0
3.5 g
3.0 3;
2.5 O
2.0 |
1.5 c
1.0 m
CM"
0.5 °
0.0
1 3 5 7 9 11 13 15 17 19 21 23 25
2 4 6 8 10 12 14 16 18 20 22 24 26
Run No.
"•" Volumetric Flow
~~*~ 02, Economizer Outlet
Figure 2-8. Process stability at Lake Hubbard (high load), illustrated by
volumetric flow (determined by the baseline Autoprobe, 16-
point, straight-up) and O2 concentration at the economizer
outlet.
2-7
-------
w
^ t~
O
0)
E
1
1600 4n
1580
1560
1540
1520
1500
1480
1460
1440
__ -
- ~^+++^^^~^+~^* /W^ "
_•«*• )1_
~ ••• "^ • I i ^*
1420 (_ ~ i ! • •
1400 U \ I J m
1380 L \i
1360 L \ I
1340 |- |
1320 U
nnn i i i i i i i i i i i i i : i i i i i t i i i i i i i i i i i i i i : i i i
3.5
3.0
•} C
£~.^J
2.0
1.5
1.0
0.5
0.0
1 3 5 7 9 11 13 15 17 19 21 23 25B 27 29 31 33 35 37
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38
Run No.
~*~ Volumetric Flow
-*- O2, Economizer Outlet
C?
'S
O
o>
.N
C
LU
CM"
0
Figure 2-9. Process stability at Homer City, illustrated by volumetric flow (determined
by the baseline Autoprobe, 16-point, straight-up) and O2 concentration at
the economizer outlet.
At Homer City, the load and flow values for Run 25B3 (Matrix B testing of the French probe) were
notably lower than the values for the other runs (Figures 2-6 and 2-9). The load and flow values for
Run 25B were 6.29% and 6.62% lower than the corresponding values for Run 1. Across all other
runs, the maximum deviations of load and flow values from the values in the first run were 1.30%
(Run 28) and 3.15% (Run 9), respectively. For this reason, Run 25B was excluded from all
subsequent analyses.
2.2 VERIFICATION OF PROBE CALIBRATIONS
All copies of all probes used in the field were calibrated before the start of field testing (designated
as pre-test calibrations) and after all testing was completed (designated as post-test calibrations). In
addition, several types of probes were recalibrated after the start of testing due to physical changes
made to these probes either during or after the first field test. The Type S probes were recalibrated
after their thermocouples were modified because the initial construction made temperature
stabilization unacceptably slow. The original spherical probes were recalibrated after being repaired
to eliminate thermal stress fractures sustained during the Matrix A tests (Runs 1 -4) in the first field
test at DeCordova (see Section 2.2.1). A replacement DAT probe was also initially calibrated
between the second and third field test, when the original DAT probe developed an irreparable leak.
Due to a transmission line problem, the plant had to reduce load by approximately 323 MWe during Run 25. The run was
invalidated due to the within-run load reduction. The invalid run was designated as Run 25 A. The run was then repeated and
designated as Run 25B.
2-8
-------
All of the field test data reported in this document were derived using the most recent calibration
performed prior to each field test, that is, either the initial pre-test calibrations or the recalibrations.
The NCSU post-test calibrations were used to determine if the probe calibrations changed over the
course of testing, and, if so, the extent to which the changes affected test results. Independent
calibrations of the test probes were performed by the National Institute of Standards and Technology
(NIST) after the NCSU post-test calibrations were completed. The NIST calibrations were used to
independently verify the NCSU calibrations and to ensure that no errors were introduced due to the
design of the Merrill Wind Tunnel, the equipment used in the NCSU testing, or the procedures
followed by testing personnel.
The NIST calibrations were conducted in the NIST Dual-Test Section Wind Tunnel in Gaithersburg,
Maryland, using the same written protocol used during the NCSU calibrations4. During the NIST
calibrations, the probe pressures were read by three types of differential pressure measuring devices
that were connected in parallel: an inclined manometer, an electronic manometer, and a set of
magnehelic gauges. All the NCSU measurements were obtained using magnehelic gauges.
The tests at NCSU and NIST did not include the calibration of a "control" probe, that is, one that had
not been involved in the field tests. Therefore, it is not possible to distinguish the extent of the
change in calibration coefficients attributable to field use as opposed to variability inherent in the
performance of the calibrations.
Details of the probe calibration procedures and data analyses, including side-by-side comparisons
of the NCSU pre-test, NCSU post-test, and NIST calibration factors, are presented in Appendices
ClandC2.
2.2.1 Calibration Comparison: Pre-Test NCSU to Post-Test NCSU
One- and Two-Dimensional Probes
Calibration coefficients for the Prandtl, French, Type S, Autoprobe Type S, and modified Kiel
probes were derived at wind tunnel velocities of 30, 60, and 90 ft/sec. The field test results were
computed for these probes using pre-test Cp values at 60 ft/sec. Review of the pre- and post-test
probe calibration coefficients derived at 60 ft/sec, which are presented in Table 2-1, show that all
the post-test Cp values were within ±1.5% of the pre-test values. Method 2 contains a precision
specification limiting the deviation between the three individual Cp values (obtained from the three
repetitions at a single velocity) and the average Cp for each probe to 0.01 or less. All probes satisfied
this specification.
Summary tables that present the calibration results for each probe at each tested velocity are
presented in Appendix C1. The largest differences between pre-test and post-test Cp values occurred
at 30 ft/sec. For all probes, the post-test 30 ft/sec Cp values were higher than the pre-test coefficients,
ranging from 0.4% for Prandtl probe PR-2 to 5.1% for Type S probe S-l 1. Pre- and post-test Cp
values obtained at 60 and 90 ft/sec were generally consistent. Even with the comparatively large
changes observed at 30 ft/sec, Cp values averaged over all three velocities changed by less than 1.5%
between pre- and post-test calibrations.
The Cadmus Group, Inc. 1997. "Flow Reference Method Testing and Analysis, Test Plan: Velocity Probe Calibrations at the
NIST Wind Tunnel," prepared for the U.S. Environmental Protection Agency under Contract No. 68-W6-0050, Work
Assignment No. 0007AA-23, 16 pgs. plus appendices.
2-9
-------
Table 2-1. One- and Two-Dimensional Probe Calibration Coefficients Derived at 60
ft/sec Nominal Wind Tunnel Velocity
Probe
Type
Prandtl
French
TypeS
Autoprobe
TypeS
Modified Kiel
Probe
ID
PR-1
PR-2
PR-3
PR-4
FR-1
FR-2
FR-3
FR-4
S-10
S-ll
S-12
S-13
026
027
028
033
036
MK-1
MK-2
MK-3
MK-4
NCSU Pre-Test
Calibrations (March-
May 1997) Cp
0.976
0.970
0.976
0.976
0.711
0.704
0.710
0.708
0.802
0.803
0.801
0.801
0.785
0.788
0.783
0.789
0.770
0.736
0.736
0.736
0.736
NCSU Post-Test
Calibrations (October-
November 1997) Cp
0.982
0.970
0.972
0.970
0.705
0.699
0.711
0.699
0.790
0.797
0.796
0.797
0.784
0.785
0.779
0.780
0.775
0.734
0.747
0.746
0.744
% Difference*
0.6
0.0
-0.4
-0.6
-0.8
-0.7
0.1
-1.3
-1.5
-0.7
-0.6
-0.5
-0.1
-0.4
-0.5
-1.1
0.6
-0.3
1.5
1.4
1.1
a % Difference = [(CPNCSU^ - C[)NCSU'") / CpNCSUl>rc ] x 100
The magnitudes of the changes observed at 30 ft/sec are within the error expected for magnehelic
gauges such as those used in the NCSU probe calibrations. According to the NCSU principal
investigator, a 2% to 3% error could be expected in gauge readings, given the lower pressures
generated at the low tunnel speeds (as low as 0.2 in. H2O measured by the Prandtl probe).
Three-Dimensional Probes
Comparisons of calibration data for the 3-D probes are made on a velocity basis. To facilitate these
comparisons, a pre-test velocity data set was generated for each probe using the pre-test pressure data
and the resulting set of calibration values (F, and F2 at each pitch angle). A velocity was computed
for each probe pitch angle at each wind tunnel velocity setting. Post-test velocities were then
computed using the post-test F, and F2 calibration records that were applied to the same pre-test
point-by-point pressure data. DAT and spherical probe calibration data comparisons are presented
in Appendix Cl.
DAT Probes. For three of the four DAT probes, differences in velocities calculated using pre- and
post-test NCSU calibrations were less than 2% in the -20° to +20° pitch angle range. In the pitch
angle range of -10 ° and +10°, which is comparable to the range of pitch angles measured at the three
2-10
-------
utility stacks during this field study, the differences in velocity were less than 1 %. For reasons that
are unclear, the velocity differences ranged from 3.35% to 4.18% within the ±10° pitch angle range
for the Cadmus DAT probe (3D-3).
Spherical Probes. The first set of spherical probes (copies SPH-1 through SPH-4) was calibrated
on four occasions in 1997: in April, before the first field test; in June, after repairs were made to the
probes and before performing the second field test; in July after the second field test; and in October,
following the completion of the third test. After the first set of probes (referred to collectively as
Spherical A) was repaired, four new spherical probes (designated as copies SPH-5 through SPH-8
and referred to collectively as Spherical B) were purchased for use during the third field test. These
Spherical B probes were calibrated in August before the third field test and again in October,
following the third field test.
Because the April calibrations on the Spherical A probes were negated by the repairs, the June results
were used as the pre-test calibrations for the Spherical A probes; the July calibrations for the
Spherical A probes are referred to as "intermediate" calibrations. The August results for the four
Spherical B probes were used as the pre-test calibrations for those probes. For both sets of probes,
the October results were used as the post-test calibrations.
A review of all four sets of NCSU calibrations of the Spherical A probes showed two distinct groups.
First, the final (October) calibration records compared well with the initial calibration, even though
it was believed that the probe repairs would have changed the initial probe calibration; second, the
June and July calibrations compared well. The differences between these two pairs of calibration
records are approximately 2% to 3% within the-10° to+10° pitch angle range; the differences
increase at pitch angles beyond ±10°. The results of the four calibrations show that the calibration
values may vary by up to 5% over all pitch angles.
The Spherical B probe copies were calibrated only twice, once before and once after the Homer City
field test. The differences in these pre- and post-test calibrations were greater than those determined
for the Spherical A probe copies. On average, the change in calibration was approximately 4% to
5% within the -10° to +10° pitch angle range.
2.2.2 Calibration Comparison: NISTtoNCSU
In the probe calibrations performed at NIST, excellent agreement was achieved between the resulting
calibration factors generated using the electronic manometer, inclined manometer, and magnehelic
gauges; the differences, in most cases, were less than 1%. Because of this agreement, the NIST
calibrations used in all comparisons presented below are based exclusively on the inclined
manometer readings. The complete presentation of calibration factors generated using the three
pressure devices is included in Appendix C2.
One and Two-Dimensional Probes
The calibration procedures performed at NIST were identical to those performed at NCSU.
Calibration coefficients for the Prandtl, French, Type S, Autoprobe Type S, and modified Kiel
probes were derived at nominal wind tunnel velocities of 30,60, and 90 ft/sec, and the resulting Cp
values were calculated. Review of the NIST and NCSU post-test probe calibration coefficients
derived at 60 ft/sec, which are presented in Table 2-2, show that the NCSU post-test Cp values were
2-11
-------
within ±2.2% of the NIST values. NIST Cp values generally were higher than the corresponding
NCSU post-test values; percent differences between the NIST andNCSU values varied slightly from
probe to probe.
Summary tables of the calibration results for each probe at each tested velocity are presented in
Appendix C2.
Table 2-2. One- and Two-Dimensional Probe Calibration Coefficients Derived
at 60 ft/sec Nominal Wind Tunnel Velocity
Probe
Type
Prandtl
French
TypeS
Autoprobe
TypeS
Modified Kiel
Probe
ID
PR-1
PR-2
PR-3
PR-4
FR-1
FR-2
FR-3
FR-4
S-10
S-ll
S-12
S-13
026
027
028
033
036
MK-1
MK-2
MK-3
MK-4
NCSU Post-Test
Calibrations
(October-
November 1997) Cp
0.982
0.970
0.972
0.970
0.705
0.699
0.711
0.699
0.790
0.797
0.796
0.797
0.784
0.785
0.779
0.780
0.775
0.734
0.747
0.746
0.744
NIST
Calibrations
CP
0.9970
0.9885
0.9890
0.9887
0.7093
0.7093
0.7100
0.7094
0.8022
0.8051
0.7909
0.8088
0.7941
0.7931
0.7963
0.7921
0.7852
0.7483
0.7485
0.7508
0.7471
% Difference"
-1.5
-1.9
-1.7
-1.9
-0.6
-1.5
0.1
-1.5
-1.5
-1.0
0.6
-1.5
-1.3
-1.0
-2.2
-1.5
-1.3
-1.9
-0.2
-0.6
-0.4
a % Difference = [(CpNCSUpost - CPNIST) / C.NIST] * 100
Three-Dimensional Probes
DA T Probes. Comparisons of the NIST and NCSU post-test calibration results for 3-D probes were
made on a velocity basis, as described in the earlier subsection on comparisons of NCSU pre- and
post-test calibration results for 3-D probes. For the NIST vs. NCSU post-test comparisons, a
baseline velocity data set was generated for each probe using the NCSU pre-test pressure data and
calibration values (Fj and F2 at each pitch angle). The NIST and NCSU post-test calibration values
for each probe were applied to the NCSU pre-test pressure data for that probe and the resulting
velocities were compared.
A comparison of calculated velocities within the -10° to +10° pitch angle range shows good
agreement between the NIST and post-test NCSU calibrations. Within this pitch range, the average
2-12
-------
percent difference between the NCSU and NIST calculated velocities over the three nominal wind
tunnel velocity settings (i.e., 30, 60, and 90 ft/sec) was 1.7% or less for each of the tested DAT
probes (Table 2-3).
Table 2-3. Comparisons of Calculated Velocities Using the NIST and
NCSU Post-test Calibrations of DAT Probes
DAT
Probe ID
E-DAT
K-DAT
T-DAT
3D-3
Average % Difference": NIST vs. NCSU Post-Test Calibrations
0°,±5°,±10° Pitch
-0.3
-0.6
-1.7
-1.5
±15° and ±20° Pitch
-2.0
-1.0
-3.0
-3.3
±25° and ±30° Pitch
-3.0
-1.6
-2.4
-3.8
a % Difference = [(NCSU"0" - NIST) / NIST] x 100.
Beyond the ±10° pitch range, the differences between the velocities derived using the NCSU and
NIST calibration curves are greater. Table 2-3 shows the average differences between NIST and the
NCSU calibrations for tests conducted at the following groups of pitch settings: (1) 0°, ±5°, and
±10°; (2) ±15° and ±20°; and (3) ±25° and ±30°. For all DAT probes, the average percent
difference between the NIST and NCSU post-test velocities was greater for pitch angles outside the
range of ±10° than for pitch angles within the range of ±10°. Because the pitch angles found at all
three sites were generally in the range of ±10°, only the results in the ±10° pitch range are relevant
to this study. Considering the research-grade quality of both the NCSU and NIST wind tunnels, the
high level of quality control exercised during both calibrations, and the professional experience of
the test personnel, the results shown in Table 2-3 suggest that aerodynamic factors may make
obtaining stable calibration curves at pitch angle settings beyond ±10° increasingly difficult.
Spherical Probes. A comparison of the calculated velocities for the NIST and NCSU post-test
spherical probe calibrations (Table 2-4) shows that, within the -10° and +10° pitch angle range, the
calculated NCSU velocity for one probe (SPH-1) was low relative to NIST (-1.2%). For all other
probes, the NCSU calculated velocities were 1.9% to 6.4% higher than those obtained by NIST.
Unlike the DAT probes, the differences between the velocities derived using the NCSU and NIST
calibration curves did not change significantly at settings outside the ±10° pitch range (Table 2-4).
Table 2-4 also shows the average differences between NIST and the NCSU calibrations for each
probe for tests conducted at the following groups of pitch angles: (1) 0°, ±5°, and ±10°; (2) ±15°
and ±20°; and (3) ±25° and ±30°.
Summary tables of the calibration results for each DAT and spherical probe at each tested velocity
are presented in Appendix C2.
2-13
-------
Table 2-4. Comparisons of Calculated Velocities Using the NIST and
NCSU Post-test Calibrations of Spherical Probes
Spherical
Probe ID
SPH-1
SPH-2
SPH-3
SPH-4
SPH-5
SPH-6
SPH-7
SPH-8
Average % Difference1: NIST vs. NCSU Post-Test Calibrations
0°, ±5°, ±10° Pitch
-1.2
4.4
3.4
1.9
6.1
4.6
5.7
6.4
±15° and ±20° Pitch
-1.7
3.6
2.5
2.0
5.3
3.3
4.1
5.8
±25° and ±30 "Pitch
-0.2
3.7
3.0
2.3
4.3
2.8
2.9
5.1
a % Difference = [(NCSU"051 - NIST) / NIST] x 100.
2.3 ENGINEERING METHOD ERROR ANALYSIS
For this study, two types of engineering methods were used to estimate volumetric flow: (1) four
methods based on combustion stoichiometry (O2 F-factor, CO2 F-factor, MMBtu, and ASME PTC
4.1) and (2) one method based on conservation of mass and energy (BTCE). Two of the four
stoichiometric methods use F-factors. F-factor methods of estimating flue gas velocity are based on
estimating the ratio of flue gas generated from combustion of fuel to the gross calorific value of the
fuel burned. The ratio, which is specific to the type of fuel burned, is expressed in terms of standard
cubic feet (scf) per million Btu (MMBtu). The dry F-factor (Fd), used in the O2 F-factor method,
includes all components of combustion less water, while the carbon F-factor (Fc), used in the CO2
F-factor method, includes only CO2. In the O2 F-factor method, Fd is used in combination with
measured values of flue gas O2 concentration and total heat input to the boiler to estimate the
volumetric flow rate of the flue gas. In the CO2 F-factor method, Fc is used with measurements of
CO2 concentration in the flue gas and total heat input to the boiler to estimate volumetric flow.
The MMBtu method is the third stoichiometric method. This method is based on the principle that
the weight of air required to bum a unit weight of fuel is proportional to the amount of heat input.
Several steps are involved in the calculation procedure, including deriving the weights of wet and
dry combustion air and wet gaseous products of combustion, which are expressed in units of MMBtu
fired. The weight of gaseous combustion products per MMBtu is multiplied by the total heat input
to the boiler in MMBtu to determine the flue gas mass flow rate, which is then divided by the density
of the flue gas to derive the flue gas volumetric flow rate.
In the fourth stoichiometric method, which is based on procedures for determining plant heat rate
described in the American Society for Mechanical Engineers Power Test Code (ASME PTC) 4.1,
the flue gas mass flow rate is calculated using the as-fired (AF) fuel mass flow rate and the
calculated weight of wet flue gas per pound of AF fuel. The weight of wet flue gas per pound of AF
fuel is calculated from the weight of dry flue gas per pound of AF fuel, which is based on measured
concentrations of CO2,02, and CO in the flue gas and stoichiometric combustion of fuel carbon and
sulfur, along with the weight of flue gas moisture per pound of AF fuel, which takes into account
moisture in the AF fuel and combustion air. The weight of ash per pound of AF fuel and ash heating
2-14
-------
value are also taken into account in determining the weight of dry flue gas per pound of AF fuel.
The flue gas volumetric flow rate is determined by multiplying the flue gas mass flow rate by the flue
gas density, which is based on measured concentrations of H2O, CO2, O2, and CO in the flue gas.
The Boiler Turbine Cycle Efficiency (BTCE) method can be used to estimate the flue gas volumetric
flow rate generated from fossil fuel combustion based on conservation of mass and energy. Unlike
the four methods described above, which require quantitative data on fuel feed rate, the BTCE
method does not require measurement of fuel feed rate. It does, however, require approximately 20
measured plant parameters, such as economizer excess O2 concentration, fuel composition, and
steam, air, and fuel temperatures, as well as calculated performance data, such as turbine cycle heat
rate and air preheater leakage rate. These data are then used as inputs to solve a series of complex
equations based on conservation of mass and energy for the combustion side of the power plant.
A formal quantitative error analysis, called first-order error analysis, was implemented for each
engineering method evaluated at the three test sites. First-order error analysis answers the question:
"What is the total uncertainty in any single estimate of volumetric flow based on a specific
engineering calculation?" The analysis was performed by the Lehigh University ERC before the field
studies began, using data from historical studies and literature values, and again after the field tests
were completed, using data generated during the course of the field testing. The results of the
analyses are presented in this section.
It is important to note that error analysis (also referred to as uncertainty analysis) provides a measure
of the variability (i.e., uncertainty) inherent in a method, not its accuracy. Error analysis reveals the
variation that can be expected in a method's results based on the variation in its input parameters,
irrespective of whether the method is a good or poor indicator of the true flow value. Accepting the
assumptions of the method, error analysis quantifies the variability that can be expected in the
method's results.
The first-order error analysis consists of several steps. First, an equation providing an approximation
(to first order) of the error in a single calculation of volumetric flow is derived. The first-order
equation propagates the error in each term of the engineering method equation into an error in the
calculation of volumetric flow. Assuming independence among the input parameters for each of the
engineering method calculations, the overall (resulting) uncertainty in the calculated flow rate can
be determined from the uncertainties in the primary measurement parameters (such as temperature
and pressure) using the following approach:
U(R) = {[8,U(Xl)f + [02U(x2)J2 + + [6nU(xn)]2}°-5 Eq. 2-1
where:
U(R) = resulting coefficient of variation (C V) in calculated volumetric or mass flow rate of
flue gas;
Ol = sensitivity coefficient for the i-th parameter, i = 1 to n;
U(x,) = measurement uncertainty in the i-th parameter as represented by the sampling
variance of the parameter, i = 1 to n;
;c, = the i-th parameter, i = 1 to n; and
n = the total number of parameters.
2-15
-------
Sensitivity coefficients for the stoichiometry-based engineering methods were determined by
differentiation, while sensitivity testing of the engineering equation was used to determine the
sensitivity coefficients for the BTCE method. This analysis was performed using the HEATRT
computer code developed by the ERC at Lehigh University. Tables C3-1 and C3-2 in Appendix C3
present the input data used in the first-order error equation for each engineering method at each field
test site.
After solving the first-order error equation, the uncertainty in the calculation of a single volumetric
flow value for each of the engineering equations was generated. The results are presented in Table
2-5 for both the pre- and post-test uncertainty analysis. The pre-test uncertainties were based on data
from historical studies and literature values. The post-test uncertainties shown in this table were
derived from the data generated during the field tests. The post-test uncertainties shown in Table
2-5 are expressed as coefficients of variation, a statistical term obtained by dividing the standard
deviation of a parameter by its mean value and multiplying the result by 100 to obtain a percent
value. Plots of the volumetric flow values and the respective uncertainties associated with each
engineering method are presented in Figures C-3-1 through C-3-3 in Appendix C-3.
Examination of Table 2-5 provides the following information:
• The MMBtu method had the largest uncertainty at Homer City, while the CO2 F-factor
method had the largest uncertainty at DeCordova and Lake Hubbard.
• The coefficients of variation at the coal-fired site were two to three times larger than those
at the gas-fired sites.
• The uncertainties at the gas-fired sites were, except for the CO2 F-factor method, less than
half of those at Homer City.
Table 2-5. Pre- and Post-test Error Estimates of Engineering Methods
Flow
Calculation
Method
O2 F-Factor
CO2 F-Factor
MMBtu
ASMEPTC4.1
BTCE
Pre-Test Uncertainty Analysis
Resulting Uncertainty (%)
Natural Gas-fired"
2.5 to 3.5
5 to 6
2 to 3.5
5 to 6
3 to 3.5
Coal-firedb
2.5 to 3.5
5 to 6
2 to 3.5
5 to 6
3 to 3.5
Post-Test Uncertainty Analysis
Coefficient of Variation (%)
DeCordova
1.20
1.62
0.65
1.40
0.77
Lake Hubbard
0.52
0.68
0.40
0.49
0.35
Homer City
2.88
2.97
3.16
2.89
2.39
a In calculating the uncertainties for each engineering method at the gas-fired sites, the following uncertainties were associated
with key input parameters: 1-2% in HHV, 1-2% in M^,, 5% in O2, 5% in CO2,2-3% in HR^^, 0.1% in O2sa&, 5% in excess
air coefficient, 1% in fuel composition, 1% in Fd, 1% in Fc, and 1% in A^. (See Appendix H for definitions of terms.)
At the coal-fired site, the same uncertainties were associated with the input parameters as at the gas-fired sites. In addition, an
uncertainty of 2% Abs. was associated with loss on ignition (LOI).
2-16
-------
SECTION 3
ANALYSIS OF MATRIX A AND MATRIX B
This section analyzes the Matrix A (interprobe) and Matrix B (intraprobe) test results. Section 3.1
presents between-method comparisons using a series of analytical approaches. Section 3.2 examines
each in-stack method individually to determine whether the choice of particular probe copies and test
teams affected the results obtained. In Section 3.3, the variability of each method is examined.
Section 3.4 summarizes the results presented in Sections 3.1 and 3.2 in terms of two key analytical
factors: proximity to the central tendency of the data and the comparative variability of the methods.
Section 3.5 compares the expected uncertainty in engineering methods with the expected uncertainty
in the probe methods.
The analysis presented in this section represents the first of two rounds of analyses of the field test
data. A supplemental round of analyses was conducted in response to recommendations provided
to EPA by a panel of peer reviewers based on their review of the analyses presented in this section.
These follow-up analyses are presented in Section 4.
3.1 COMPARISON ACROSS PROBES AND METHODS
Four analytical approaches are used in the sections below. Section 3.1.1 uses descriptive statistics
and graphs to describe the flow characteristics at each site and to compare the volumetric flow results
obtained by the various methods. Section 3.1.2 presents the results of rank order analyses to reveal
patterns among the methods more clearly. Section 3.1.3 applies an analysis of variance (ANOVA)
to the Matrix A volumetric flow values to determine whether statistically significant differences can
be detected among the individual methods and classes of methods. Section 3.1.4 analyzes the
proximity of each method's measurements to the central tendency of the data to discover whether
certain methods are more likely to be good indicators of the average, long-term volumetric flow.
3.1.1 Relationships Among Probes and Methods
The following analyses use descriptive statistics and graphs to evaluate the relationships among
probes and methods in the Matrix A and Matrix B data sets. Specific analyses are presented that
provide information on the range of the volumetric flow measurements among all methods,
characterization of the range of volumetric flow measurements among in-stack methods only, and
the range of volumetric flow measurements among engineering methods only. In addition,
volumetric flow values derived from in-stack measurements are compared with those derived from
the engineering methods (see Table 3-4). The manual Type S probe straight-up measurements are
also compared with the engineering methods (see Table 3-5) to identify patterns among the methods
across the DeCordova, Lake Hubbard, and Homer City test sites.
Overall Range of Measurements
This analysis evaluates the extent of the differences in volumetric flow measurements among the
methods tested in this study. Only those volumetric flow values obtained during stable load
conditions were used in the analysis (e.g., Run 25B at Homer City was dropped from the analysis,
as described in Section 2.1.4). The average volumetric flow value for each method tested in the
experimental design was calculated. The methods with the largest and smallest average values are
shown in Table 3-1. In addition to the average values for these methods, the difference between the
maximum and minimum average value is presented to show the range of volumetric flow
3-1
-------
measurements at a specific site. Because the flow is stable across the experiment (see Section 2.1.4),
differences in volumetric flows can be attributed to inherent differences among methods. To
simplify an examination of the range across different sites, the ratio of the range (maximum minus
minimum) to the average of all methods was computed and is displayed in Table 3-1.
Table 3-1. Overall Range in Flow Measurements
Max (wscfm)
Min (wscfm)
Difference
DeCordova
Avg. Value
1,631,513
1,528,137
103,176
(6.58%)c
Method
S Straight-up
O2 F-factor
Lake Hubbard"
Avg. Value
1,157,629
1,036,472
121,157
(11.03%)°
Method
Modified
Kielb
Spherical
Homer City
Avg. Value
1,466,549
1,359,861
106,688
(7.56%)°
Method
S Straight-up
French
High-load runs only.
The modified Kiel probe produced the highest measurement of flow when measurements taken only under high-load operation
were included. When low-load measurements were also included, the Type S probe straight-up gave the highest average
measured flow; see Tables 3-6 through 3-8.
Ratio of difference relative to average of all in-stack and engineering methods, expressed as a percent.
Table 3-1 shows that the range of average flow measurements was greatest at Lake Hubbard and
least at DeCordova. At two of the three sites, the maximum mean difference was produced by the
manual Type S probe operated in straight-up mode. At Lake Hubbard (under high-load operating
conditions only), the maximum average value was produced by the modified Kiel probe. At two
sites, the lowest measurements were produced by two new probes being evaluated in this study, the
French and spherical probes.
In-stack methods characterization
In Table 3-2, the range of volumetric flow values is characterized for in-stack methods only. The
objective was to identify which particular methods defined the extremes of the range. As before,
average volumetric flow values were calculated for each in-stack method with Run 25B at Homer
City dropped from the calculation.
Table 3-2. Range in Flow Measurements for In-stack Methods
Max (wscfm)
Min (wscfm)
Difference
DeCordova
Avg. Value
1,631,513
1,552,637
78,876
(4.99%)°
Method
S Straight-up
BLAP 48-point
yaw-nulled
Lake Hubbard"
Avg. Value
1,157,629
1,036,472
121,157
(11.01%)c
Method
Modified
Kielb
Spherical
Homer City
Avg. Value
1,466,549
1,359,861
106,688
(7.53%)c
Method
S Straight-up
French
High-load runs only.
The modified Kiel probe produced the highest measurement of flow when measurements taken only under high-load operation
were included. When low-load measurements were included, the Type S probe straight-up gave the highest average measured
flow; see Tables 3-6 through 3-8.
Ratio of difference relative to average of all in-stack methods, expressed as a percent.
3-2
-------
Table 3-2 shows that, at Lake Hubbard and Homer City, the in-stack measurements (by the spherical
and French probes, respectively) defined the lower extremes in the average volumetric flow. At
DeCordova, the lowest mean average volumetric flow was obtained from the baseline Autoprobe
48-point, yaw-nulled. The highest average in-stack methods were the manual Type S straight-up at
DeCordova and Homer City and the modified Kiel at Lake Hubbard.
Engineering methods characterization
The results of performing the same calculations only on the engineering method data are presented
in Table 3-3. Not surprisingly, the difference among engineering methods was greatest at the coal-
fired facility, Homer City, where tracking combustion factors is most difficult.
Table 3-3. Range in Flow Measurements for Engineering Methods
Max (wscfm)
Min (wscfm)
Difference
DeCordova
Avg. Value
1,547,958
1,528,137
19,821
(1.29%)b
Method
MMBtu
O2 F-factor
Lake Hubbard8
Avg. Value
1,109,560
1,077,270
32,290
(2.95%)b
Method
ASME
BTCE
Homer City
Avg. Value
1,420,873
1,366,783
54,089
(3.87%)b
Method
CO2 F-factor
BTCE
a High-load runs only.
b Ratio of difference relative to average of all engineering methods, expressed as a percent.
Comparison of in-stack and engineering methods
In-stack and engineering volumetric flow methods are compared in Table 3-4. The numbers of runs
analyzed are shown in parentheses. First, the average volumetric flow measurements of all in-stack
methods and of all engineering methods were computed separately. The difference in these values
was computed, and the ratio of the difference to the average engineering value was calculated. The
resulting value is the percent difference in the two classes relative to the engineering method average
volumetric flow. Second, the same calculation was performed using only the manual Type S
straight-up method in place of all in-stack methods, and only the engineering values for the runs
when the manual Type S probe was tested were used to compute the average. The Type S straight-
up method is the current Method 2 approach.
Table 3-4. In-stack and Engineering Methods Comparison'
Comparison
In-stack Avg. vs. Engineering Avg.
S-straight Avg. vs. Engineering Avg.
DeCordova
2.50% (32)
7. 10% (8)
Lake Hubbard
0.26% (20)
5.44% (8)
Homer City
1.41% (31)
5.43% (8)
Number of runs analyzed is shown in parentheses.
The difference between the engineering methods and the manual Type S probe straight-up is further
shown in Table 3-5. Table 3-5 presents the average and maximum percent difference of the manual
Type S probe's volumetric flow measurements from the volumetric flow values derived by each of
the engineering methods. These statistics were calculated using all Matrix A and Matrix B runs in
which a Type S probe was tested. The number of tests used in the calculations is also shown.
3-3
-------
Table 3-5. Average Difference of Manual Type S Probe Straight-up
from Engineering Methods for Matrix A and Matrix B Runs
Method
O2 F-Factor
CO2 F-Factor
MMBtu
ASMEPTC4.1
BTCE
DeCordova
20 Tests
Average
7.57%
7.40%
6.28%
7.02%
6.17%
Maximum
10.54%
10.14%
9.26%
9.82%
8.94%
Lake Hubbard
(High Load)
21 Tests
Average
5.56%
3.54%
4.33%
3.15%
6.40%
Maximum
11.95%
9.51%
10.11%
8.80%
11.26%
Homer City
20 Tests
Average
3.28%
3.35%
6.02%
5.83%
9.42%
Maximum
9.61%
9.92%
12.65%
12.49%
13.79%
Table 3-5 shows that, depending on the engineering method used, the average difference ranged from
6.17% to 7.57% at DeCordova, 3.15% to 6.40% at Lake Hubbard, and 3.28% to 9.42% at Homer
City. If the BTCE results at Homer City are excluded, the data indicate that the difference was
greatest at the site where the flow was closest to axial (DeCordova) and lower at the other two sites.
This observation is inconsistent with the general perception that non-axial flow heightens the
disparity between manual Type S straight-up measurements and engineering method calculations.
Consistent patterns across sites
Using the data in Tables 3-6 through 3-8, patterns among the methods were identified across
DeCordova, Lake Hubbard, and Homer City. To facilitate comparisons across sites with
significantly different volumetric flow values, the flow values produced by each tested probe/
procedure combination and engineering method were normalized relative to measurements produced
by a baseline method.
The baseline chosen for the current analysis was the set of measurements taken by the 16-point
baseline Type S Autoprobes operated in straight-up mode, that is, without yaw-nulling.
(Comparisons using two other methods as baselines, i.e., the BTCE and MMBtu methods, appear
in the series 13 and 14 tables included in the site data reports for the DeCordova, Lake Hubbard, and
Homer City field tests.) As was noted in the original test plans for the field studies, comparison of
measurements to a particular "baseline" is simply a tool for analyzing and displaying the relative
magnitudes and variability of the volumetric flow obtained by each of the tested methods5. Choosing
the 16-point straight-up Autoprobe measurements as a common denominator for comparing the
various methods does not imply that its measurements represent or approximate the true flow rate
any more accurately than the other tested methods. In part, this set of data was chosen because the
baseline Autoprobes were operated during all runs at all three sites and are allowed to be used in
current Method 2.
See, for example, The Cadmus Group, Inc., 1997, "Flow Reference Method Testing and Analysis: Field Test Plan, DeCordova
Steam Electric Station," EPA/430/R-97-024.
3-4
-------
Table 3-6. Summary of Volumetric Flow Results—DeCordova
Probe/Method
DAT
Type S Straight-up
Type S Yaw-nulled
Modified Kiel
Spherical
Prandtl
French
Manual Autoprobe 16-point Straight-up
Manual Autoprobe 16-point Yaw-nulled
Baseline Autoprobes 16-point Straight-up"
Baseline Autoprobes 16-point Yaw-nulled
Baseline Autoprobes 48-point Straight-up
Baseline Autoprobes 48-point Yaw-nulled
O2 F-Factor
CO2 F-Factor
MMBtu
ASMEPTC4.1
BTCE
No.
Tests
30
20
20
20
9
20
20
4
4
31
22
31
22
31
31
31
31
31
Percent Difference From Baseline"
Mean
-0.18%
4.50%
4.00%
1.76%
0.75%
3.70%
0.09%
0.48%
0.35%
0.00%
-0.14%
-0.64%
-0.75%
-2.38%
-1.51%
-1.14%
-1.21%
-1.58%
Minimum
-4.43%
1.08%
0.59%
-0.42%
-1.09%
1.89%
-3.37%
-0.03%
0.10%
0.00%
-0.45%
-0.91%
-0.97%
-4.17%
-4.64%
-2.31%
-4.20%
-2.49%
Maximum
2.86%
7.38%
7.98%
5.41%
2.81%
5.28%
2.94%
1.15%
0.60%
0.00%
0.33%
-0.49%
-0.39%
-1.04%
0.40%
-0.15%
0.36%
-0.46%
s.d.
2.03%
1.62%
1.80%
1.37%
1.35%
0.92%
1.68%
0.50%
0.21%
0.00%
0.19%
0.11%
0.16%
0.76%
1.32%
0.63%
1.18%
0.41%
a Baseline is baseline Autoprobes 16-point straight-up.
The box-and-whisker plots shown in Figures 3-1 through 3-6 provide additional visual
representations of the data presented in Tables 3-6 through 3-86. (Note that Figures 3-4 through 3-6
illustrate the same results as presented in Tables 3-1 through 3-3, but the scales have been
normalized.) The following observations can be made:
• At each site, the performance characteristics of the single manual Autoprobe were very
similar to those of the baseline four-Autoprobe system.
• At all three sites, the manual Type S straight-up measurements exhibited a higher average
positive percent difference from the baseline than all other methods.
• The Prandtl probe exhibited the lowest standard deviation of percent difference in
volumetric flow from the baseline of any of the in-stack methods tested at DeCordova and
Lake Hubbard, except the Autoprobes. (The Prandtl probe was not tested at the coal-fired
Homer City site, because of the possibility of plugging.)
The box-and-whisker plot is a useful tool for displaying the distribution of a data set. The plots show the median (middle
horizontal line), the 25th and 75th percentiles (lower and upper horizontal lines of the boxes), and the minimum and maximum
values (lower and upper ends of the whiskers, or lowest and highest asterisks [*]). Potential outliers are indicated by asterisks
(*) and the criterion for identifying the potential outliers is any value beyond the distance of 1.5 times the interquartile range
(75th percentile value minus the 25th percentile value) from the box.
3-5
-------
Table 3-7. Summary of Volumetric Flow Results— Lake Hubbard
Probe/Method
DAT
Type S Straight-up
Type S Yaw-nulled
Modified Kiel
Spherical
Prandtl
French
Manual Autoprobe 16-point Straight-up
Manual Autoprobe 16-point Yaw-nulled
Baseline Autoprobes 16-point Straight-up"
Baseline Autoprobes 16-point Yaw-nulled
Baseline Autoprobes 48-point Straight-up
Baseline Autoprobes 48-point Yaw-nulled
O2 F-Factor
CO2 F-Factor
MMBtu
ASMEPTC4.1
BTCE
No.
Tests
29
25
25
8
25
4
4
4
4
25
25
24
24
25
25
25
25
25
Percent Difference From Baseline"
Mean
-2.98%
3.65%
1.72%
3.11%
-7.17%
0.93%
-3.24%
0.09%
-0.93%
0.00%
-1.08%
-0.99%
-2.06%
-1.75%
-0.02%
-0.50%
0.49%
-2.31%
Minimum
-13.44%
0.98%
-3.70%
-4.24%
-38.73%
-0.30%
-7.37%
-0.71%
-1.58%
0.00%
-2.93%
-1.48%
-3.13%
-3.60%
-1.78%
-1.99%
-0.81%
-3.31%
Maximum
0.54%
7.92%
4.93%
12.04%
-1.22%
3.12%
-0.16%
0.97%
-0.29%
0.00%
0.47%
-0.18%
-0.88%
0.78%
2.69%
1.21%
2.37%
1.60%
s.d.
3.10%
1.83%
2.16%
4.71%
7.84%
1.52%
3.16%
0.75%
0.52%
0.00%
0.66%
0.36%
0.51%
1.12%
0.96%
0.78%
0.76%
0.99%
Baseline is baseline Autoprobes 16-point straight-up.
Table 3-8. Summary of Volumetric Flow Results—Homer City
Probe/Method
DAT
Type S Straight-up
Type S Yaw-nulled
Modified Kiel
Spherical A
Spherical B
French
Manual Autoprobe 16-point Straight-up
Manual Autoprobe 16-point Yaw-nulled
Baseline Autoprobes 12-point Straight-up
Baseline Autoprobes 12-point Yaw-nulled
Baseline Autoprobes 16-point Straight-up"
Baseline Autoprobes 16-point Yaw-nulled
Baseline Autoprobes 20-point Straight-up
Baseline Autoprobes 20-point Yaw-nulled
O2 F-Factor
CO2 F-Factor
MMBtu
ASMEPTC4.1
BTCE
No.
Tests
23
20
20
20
20
20
20
4
4
17
17
32
28
15
11
32
32
32
32
32
Percent Difference From Baseline"
Mean
-3.20%
3.48%
-1.12%
-2.39%
-3.88%
-4.36%
-6.29%
0.18%
-3.41%
0.27%
-4.14%
0.00%
-4.33%
-0.55%
-4.65%
-2.23%
-2.03%
-4.60%
-4.06%
-5.81%
Minimum
-6.15%
1.20%
-3.87%
-6.70%
-6.36%
-8.65%
-11.78%
0.04%
-3.57%
-1.73%
-5.64%
0.00%
-4.77%
-2.27%
-5.81%
-14.78%
-14.39%
-17.71%
-17.12%
-8.56%
Maximum
0.59%
5.98%
2.78%
1.11%
-0.15%
0.17%
-2.30%
0.45%
-3.22%
2.56%
-2.59%
0.00%
-3.68%
0.63%
-2.55%
2.18%
2.58%
-0.59%
-0.12%
-4.30%
s.d.
.83%
.23%
.67%
.99%
.70%
2.06%
2.85%
0.19%
0.18%
1.05%
0.71%
0.00%
0.33%
0.74%
0.88%
2.96%
2.90%
3.01%
2.94%
0.98%
Baseline is baseline Autoprobes 16-point straight-up.
3-6
-------
TO
K)
I
3
S
cr
65
O9
JL
S'
fB
O
•a
o
CP
•a
o
i
TO'
er
•H-
•a
Flow Percent Difference From Baseline (%)
i i i i
A U) N) i— N* K)
O 0 0 O 0 O 0
V*'
\V
V
X
\.
\^\
*\*
^
'O'
•C"
\ .
\
. .„=,
KD 1
H-^H
, ^C^H »
* * * 1 — CD — 1
inn
m
o
«n
KH.
HDH *
HD-*
HD-K
3
V*
H-»
•
a "8
^ S
» §
^ **
n Q.
S
a-
•a
•o
o
•a
Flow Percent Difference From Baseline (%
<>
•i-
\
0
«•<*>
f<,
-------
a "
3
5
CO
•L
s*
n
•
o
•o
o-
n
"O
O
3
f
7-
e
69
Flow Percent Difference From Baseline (%
1
£
a 2
e S^
5 3
3 •«
£.
S"
P
rs
cr
65
at
n
I
•a
3
n
•a
o
P
2
I
•a
Flow
•I\
\, .
V
*£*
\\
****
°'\
->,<•
%
\
Percent Difference From Baseline (%)
i i
D O O O
i i , , i i
^-H
n
ID
* KI — 1 *
+ HI}^*
DH
I — LTH
^{^M *
* *— CIZL3 1
i— irzn i
1 OH
-------
s
ofq'
re
D 2
65 «t
n
n
re
S
CP
85
C/J
2.
5"
re
o
TS
3
O"
O
s*
t
C
•a
S.
ffl
o
5
Flow Percent Difference From Baseline (%
(W
i
OJ
a 2
s ?
•1 «
H
I
31
5
a-
8!
Vl
ft
•a
"O
o
ore
or
•jt-
C
T3
65
65
s*
Flow Percent Difference From Baseline (%
<
\
-------
The yaw-nulled Type S probes (both manual and automated) always resulted in volumetric
flow values that were on average less than the Type S straight-up values.
At all sites, the range (maximum minus minimum) in the percent differences from the
baseline Autoprobe 16-point straight-up reported for the DAT probes, the yaw-nulled 16-
point Autoprobes (both manual and baseline), and the MMBtu method was less than 2.5%
(Table 3-9).
Table 3-9. DAT, Autoprobe Yaw-Nulled, and MMBtu Performance:
Percent Difference From Baseline
Probe/Method
DAT
Manual Autoprobe 16-point Yaw-nulled
Baseline Autoprobes 16-point Yaw-nulled
MMBtu
Maximum Value
Minimum Value
Range (maximum - minimum)
DeCordova
-0.18%
0.35%
-0.14%
-1.14%
0.35%
-1.14%
1.49%
Lake Hubbard"
-2.98%
-0.93%
-1.08%
-0.50%
-0.50%
-2.98%
2.48%
Homer City
-3.20%
-3.41%
-4.33%
-4.60%
-3.20%
-4.60%
1.40%
a All loads included.
3.1.2 Rank Order Analysis
Rank order tables assign a discrete integer "rank" (in this case, 1 for the lowest measurement, 2 for
the next lowest measurement, etc.) to the volumetric flow value determined by each method for each
run. Assigning discrete integer values is useful for revealing among-methods patterns that are often
difficult to detect when data are expressed in units of volumetric flow or percent difference from a
baseline.
A complete set of rank order tables for each field test is found in the three individual site data
reports. An aggregate rank order summary for the Matrix A tests across the three test sites is
presented in Table 3-10. The table was created using the following steps: (1) within each of the
eight Matrix A runs at each site, the methods were ranked from lowest to highest based on the
volumetric flow value obtained; and (2) for each method, the median rank and the range in ranks
across all runs were calculated. These values are displayed in Table 3-10; details are presented in
Appendix D.
The rank order analysis enables three questions to be examined:
1. Are there discernible patterns in the magnitudes of the volumetric flow measurements
among methods, for example, do some methods consistently display high rankings, while
others display mid-range or low rankings?
3-10
-------
2. How are the rankings affected by changes in flow conditions from site to site?
3. Are the rankings of some methods more variable than other methods?
Table 3-10. Aggregate Rank Order of Volumetric Flow at all Sites Across all
Methods for Matrix A (Runs 1-8)
Probe/Method
DAT
Type S Straight-up
Type S Yaw-nulled
Modified Kiel
Spherical Ab
Spherical Bb
Prandtf
French
Baseline Autoprobes 12-point Straight-upd
Baseline Autoprobes 12-point Yaw-nulledd
Manual Autoprobe 16-point Straight-up
Manual Autoprobe 16-point Yaw-nulled
Baseline Autoprobes 16-point Straight-up
Baseline Autoprobes 16-point Yaw-nulled
Baseline Autoprobes 48-point Straight-upe
Baseline Autoprobes 48-point Yaw-nulled6
O2 F-Factor
CO2 F-Factor
MMBtu
ASMEPTC4.1
BTCE
DeCordova
Rank
Median
10.5
13.0
12.5
13.0
11.5
-
14.0
13.0
-
-
11.5
11.0
9.0
9.0
7.0
6.5
1.5
2.0
4.5
3.0
5.0
Rank
Range"
12
2
4
2
2
-
0
7
-
-
4
2
2
3
2
2
2
2
3
2
2
Lake Hubbard
Rank
Median
6.0
13.0
12.0
14.0
1.0
-
13.0
1.5
-
-
11.5
7.0
11.0
6.5
7.0
4.0
3.0
9.5
7.5
11.5
2.0
Rank
Range"
11
1
11
2
1
-
2
10
-
-
5
3
4
4
4
2
3
5
3
3
3
Homer City
Rank
Median
6.5
14.0
11.0
9.0
5.5
6.0
-
2.5
13.0
4.5
13.0
6.0
12.0
4.0
-
-
10.0
10.0
5.0
6.0
1.5
Rank
Range"
9
0
6
8
4
7
-
7
3
8
1
3
2
3
-
-
2
3
7
6
2
a Rank range is the difference between the largest and smallest rank.
b A = spherical probe copies SPH-1, SPH-2, SPH-3, and SPH-4; see Section 2.2.1.
B = spherical probe copies SPH-5, SPH-6, SPH-7, and SPH-8; see Section 2.2.1.
c Prandtl probes were not tested at Homer City because the 0.035-in. I.D. static pressure ports could become clogged
with particulate matter at this site.
Twelve-point traverses were performed by the baseline Autoprobes at Homer City only, for comparison with results
for 16- and 20-point traverses.
e At Homer City, 20-point traverses (results not shown in table) instead of 48-point traverses were performed by the
baseline Autoprobes for comparison with results for 12- and 16-point traverses.
3-11
-------
Examination of Table 3-10 provides the following observations:
• The DAT probe showed the largest range of ranks at all three sites, indicating a large
variance in the run-by-run observations.
• The engineering methods had low ranks at DeCordova and low-to-moderate ranks at Lake
Hubbard and Homer City. This observation indicates that the engineering methods
provided the lowest volumetric flow values of all methods at the site with axial flow
(DeCordova).
• The French probe showed the greatest change in ranking from site to site, providing a
relatively high rank at DeCordova, but very low ranks at Lake Hubbard and Homer City.
• The baseline Autoprobe had generally moderate ranks at all sites, indicating that these
methods provided measurements close to the overall median flow value.
• The Prandtl probe provided high readings relative to the other probes at the two sites where
it was tested.
• The manual Type S (straight-up and yaw-nulled) probes displayed high ranks across all
three sites, indicating a consistently high reading relative to the other probes.
The relatively high rank orders (10.5 and 11.5) for the 3-D probes at the axial-flow site are not
surprising, because at DeCordova these probes function essentially as "straight-up," manual methods
in axial flow conditions.
The low-to-moderate rank orders (6.0 and 6.5) for the DAT probes at Lake Hubbard and Homer City
are as expected. The 3-D probes would be expected to provide lower results at these sites than the
1 -D and 2-D probes, because significant yaw and/or pitch angles were present at both sites. The rank
orders of 5.5 and 6.0 for the two sets of spherical probes tested at Homer City are also expected. The
rank order of 1.0 for the spherical probe at Lake Hubbard, however, is not expected and implies that
the spherical probe data from that site are suspect.
At Lake Hubbard and Homer City, where significant yaw angles were present, the rank orders of
the 12- and 16-point straight-up manual and baseline Autoprobes were relatively high, while the rank
orders of the yaw-nulled Autoprobes were low to moderate. This observation is also as expected.
The rank order of 9.0 for the yaw-compensated modified Kiel probe at Homer City is as expected.
However, the rank order of 14.0 for the probe at Lake Hubbard is surprising. Although the average
yaw angles at Lake Hubbard were not extreme (6° to 7°), one would not expect yaw-compensated
readings with the Kiel probe to be the highest readings among all of the methods.
3.1.3 Analysis of Variance of Matrix A Data
Discussion of Analysis of Variance Techniques
Analysis of variance (ANOVA) is a statistical method for indicating whether the mean values
produced by different groupings (i.e., methods, probe copies, test teams) in this field test are
3-12
-------
important predictors of volumetric flow. Multiple comparison methods (such as Duncan's Multiple
Range Test), performed in conjunction with the ANOVA, can be used to determine which specific
methods (or probe copies or test teams) produce results that statistically differ from one another.
In the first step of performing an ANOVA, a statistical model is created that is consistent with the
experimental design. (See Appendix E for a more detailed description of the model.) The model,
which consists of several terms, is fit to the data. Each term represents a different grouping.
Measures of the relative amount of variation (called mean squares) associated with each grouping
are calculated and compared to the remaining or unexplained variation in the data (the error term).
If a mean square is large, that grouping is considered to be important in determining volumetric flow
(given a pre-specified level of confidence). For example, a large mean square associated with the
test teams term in the ANOVA model indicates that the teams themselves are important in
influencing the volumetric flow measurements.
The second step of an ANOVA is performing a multiple comparison means test. The means test
indicates which group member can be distinguished from the others (e.g., which probe copy can be
distinguished from the three other probe copies). The mathematical representations of the models
are presented in Appendix E.
Matrix A Analysis of Variance Description and Results
Several ANOVA models were applied to the Matrix A data and evaluated. For example, a model
with a single grouping ("method") was evaluated as an approach for distinguishing the relative
importance of the 16 methods as significant predictors of volumetric flow. The results of this model
were inconclusive and showed that the relationships among methods were complex, with no single
method or group of methods clearly distinguishable from the others.
To simplify the model and further the evaluations, another grouping called "class" was developed.
Methods with similar characteristics were grouped into classes. The first variation of the approach
(Variant 1) created four separate classes, one each for the 1-D, 2-D, and 3-D probes, and one for the
engineering methods. The French probe, alone, was included in a fifth class because earlier analyses
showed that the probe tended to produce lower results relative to other 1-D methods. A second set
of classes (Variant 2) consisting of only the in-stack methods with the automated probes separated
from their manual counterparts was also created. In Variant 2, the classes were 1-D manual, 2-D
manual, 3-D, French, 1-D Autoprobes, and 2-D Autoprobes.
Within each class, each method had the same number of observations. The number of observations
across classes is not consistent, but a sensitivity check of the model (using simulated data where the
sample size was kept the same across classes) indicated that this issue had no effect on the results.
Therefore, the ANOVA results are not dependent on the relative sample size for any one probe.
Both models (Variant 1 and Variant 2) were shown to be significant (a=0.05). Therefore, multiple
comparison testing was performed on the class means. In multiple comparison tests, the means and
variances of the data for each class are compared in succession to determine whether a statistically
significant difference can be detected between each pair of class means. Table 3-11 presents the
results of the means tests, using Duncan's Multiple Range Test. Other multiple comparison tests
(e.g., Tukey's Multiple Range Test) produced similar results.
3-13
-------
Table 3-11. Duncan's Multiple Range Test on Probe Classes (Matrix A, a = 0.05)
Var.
la
?b
DeCordova
(ID, 2D, F, 3D) * E
lDm * (F, 3Dm, ID,, 2DJ
2Dm*(3Dm,lD1)2Da)
Lake
High Load
(ID, 2D, E) * 3D
(2Dm, lDm, 1DJ * 3Dm
Hubbard
Low Load
(ID, 2D, E) * 3D
(lDm, 2Dm, lDa, 2DJ * 3Dm
Homer City
ID * 2D, E, 3D, F
2D*F
lDm*lDa,2Dm,2Da,3Dm,F
!Da*2D.,3Dm,F
2Dm * 3Dm, F
Variant 1 Classes
ID = Type S straight-up, baseline Autoprobe straight-up, manual Autoprobe straight-up, Prandtl
2D = Type S yaw-nulled, baseline Autoprobe yaw-nulled, manual Autoprobe yaw-nulled, modified Kiel
3D = DAT, spherical
F = French
E = Engineering methods (O2 F-factor, CO2 F-factor, MMBtu, ASME PTC 4.1, BTCE)
b Variant 2 Classes
lDm = manual 1-D methods (Type S straight-up and Prandtl)
2Dm = manual 2-D methods (Type S yaw-nulled and modified Kiel)
3Dm = 3-D Autoprobes methods (DAT and spherical)
ID. = 1-D Autoprobes methods (baseline Autoprobe straight-up)
2Da = 2-D methods (baseline Autoprobe yaw-nulled)
F = French
In Variant 1 at DeCordova, the engineering methods are separable from the other methods. This
relationship, however, does not hold at the other sites. For Variant 1, the relationships among the
classes are not consistent from site to site. At the moderate-yaw/pitch (Lake Hubbard) and high-yaw
(Homer City) sites, however, the mean flow values determined by the 1-D and 3-D probes are
statistically distinguishable. Similar results are obtained for the Variant 2 classes. In Variant 2, a
statistically significant difference between the means for the manual 3-D probes and manual 1-D
probes is observed at all three sites.
3.1.4 Analysis of Central Tendency
Difference from Grand Mean
An important feature of any measurement method is the degree to which it provides values that are
close to the long-term average or central tendency of measurements made by all viable methods.
Although central tendency does not explicitly reflect the sampling variance of the method (for a
direct analysis of sampling variance, see Section 3.3), methods with high central tendency do reflect
the long-term average of all viable methods for measuring volumetric flow. In situations where the
true or accurate value of volumetric flow at any point in time is unknown, central tendency can be
used as an indicator of the closeness of a method's measurements to the true, but unknown, long-
term mean.
All methods evaluated in this study were assumed to be credible candidate indicators of volumetric
flow. The degree to which any one method differs from the long-term sample mean can be assessed
by comparison with a baseline. A reasonable baseline for this assessment is the mean of all methods
3-14
-------
(the grand mean). Those methods that consistently measure values that are much higher or lower
than the grand mean are less likely to be good indicators of the true, long-term volumetric flow, as
represented by the long-term sampling mean. In addition, those methods that consistently read much
lower than the grand mean have the greatest chance of representing a risk to the environment by
underestimating the long-term volumetric flow in the stack.
Central tendency analysis relies on the premise that the "majority rules." If on average, most
methods offer a good estimate of true flow, methods that are outside the central tendency may be
considered less accurate. Alternatively, if a particular method demonstrates superior detection
capabilities over most of the other methods, its distance from the central tendency may in fact
represent an improvement in accuracy relative to the majority of probes.
Table 3-12 presents the differences from the grand mean for all methods at each of the three sites.
For each method, the average difference across all appropriate Matrix A runs is calculated as
a - - G
r
where:
Dk = the difference from the grand mean of method k,
Mlk = volumetric flow measured at run i for method k,
G = grand mean of all methods (k) across all runs (i), and
r = number of runs in which method measurements were taken.
The grand mean used to derive the values in Table 3-12 was created by first calculating the mean of
all probes on each run, and then calculating the means of those values. This procedure reduced the
influence of sample size on the grand mean for individual probes.
The magnitude of a method's difference from the grand mean is an important consideration because
the larger the difference, the more questionable the measurements. Additionally, methods that
consistently measure low present a potential risk to the environment because they lead ultimately to
underestimates of emissions (see discussion of the field study goals in Section 1 .2). For these
reasons, a bi-directional ranking of methods is presented in Table 3-12. That is, in the columns
labeled "rank," the method with the greatest positive difference from the grand mean is given the
highest positive ranking, and the method with the greatest negative difference from the grand mean
is given the lowest negative rank. For example, at DeCordova, the method with the greatest positive
difference and highest positive ranking is the Prandtl probe with a rank of +1 1 , and the method with
the greatest negative difference and lowest negative rank is the O2 F-factor method with a rank of
-7. The table shows differences from the grand mean in units of volumetric flow (wscfm) and on
a percent basis. Different font attributes are used to indicate various degrees or categories of percent
difference.
3-15
-------
Table 3-12. Central Tendency Analysis on Matrix A Runs at Each Site
DeCordova
Grand Mean = 1,556,062.95 wscfm
Rank
11
10
9
8
7
6
5
4
3
2
1
-1
-2
-3
-4
-5
-6
-7
Method
Prandtl
Modified Kiel
Type S Str.
Type S Nulled
Spherical A
French
Man APS
ManAPN
DAT
BAP 16S
BAP 16N
BAP 48S
BAP 48N
BTCE
MMBtu
ASMEPTC4.1
CO 2 F-factor
O2 F-factor
Difference
wscfm
67095.8
56057.3
55158.5
45875.8
35686.4
28411.3
14,902.5
12,957.6
7554
4676.2
4040.6
-4355.5
-5584.2
-20682
-22658
-34888
-40931
-42508
%
4.3%
3.6%
3.5%
2.9%
2.3%
1.8%
1.0%
0.8%
0.5%
0.3%
0.3%
-0.3%
-0.4%
-1.3%
-1.5%
-2.2%
-2.6%
-2.7%
Lake Hubbard (High Load)
Grand Mean = 1,089,755.58 wscfm
Rank
11
10
9
8
7
6
5
4
3
2
1
-1
-2
-3
-4
-5
-6
-1
Method
Modified Kiel
Type S Str.
Type S Nulled
Prandtl
Man APS
ASMEPTC4.1
BAP16S
CO2 F-factor
ManAPN
MMBtu
BAP48S
BAP16N
DAT
BAP 48N
O2 F-factor
BTCE
French
Spherical A
Difference
wscfm
79070.4
65906.9
23968.2
23726.2
14430.3
13568.9
12650.9
6434.3
3227.7
2086.3
1988.5
-1226.4
-10633.5
-10777.2
-15322.1
-16809.5
-22242.7.
-133505.6
%
7.3%
6.0%
2.2%
2.2%
1.3%
1.2%
1.2%
0.6%
0.3%
0.2%
0.2%
-0.1%
-1.0%
-1.0%
-1.4%
-1.5%
-2.0%
-12.3%
Lake Hubbard (Low Load)
Grand Mean = 712,314.81 wscfm
DonL-
8
7
6
5
4
3
2
1
-1
-2
-3
-4
-5
-6
Method
Type S Str.
TypeS Nulled
ASMEPTC4.1
Modified Kiel
MMBtu
CO2 F-factor
BAP 16S
O2 F-factor
BTCE
BAP48S
BAP 16N
BAP 48N
DAT
Spherical A
Difference
wscfm
23901.4
17214.7
15306.7
14584.4
11528.7
10225.2
6451.2
5448.7
-92.3
-137.0
-2964.6
-10024.2
-16210.8
-75232.1
%
3.4%
2.4%
2.1%
2.0%
1.6%
1.4%
0.9%
0.8%
0.0%
0.0%
-0.4%
-1.4%
-2.3%
-10.6%
Homer City
Grand Mean = 1,414,649.87 wscfm
Rank
9
8
7
6
5
4
3
2
1
-1
-2
-3
-4
-5
-6
-7
-8
-9
Method
Type S Str.
Man AP S
BAP 12S
BAP 16S
TypeS Nulled
O2 F-factor
CO2 F-factor
Modified Kiel
ManAPN
DAT
ASMEPTC4.1
BAP 12N
BAP 16N
MMBtu
Spherical B
Spherical A
French
BTCE
Difference
wscfm
98056.1
51299.7
41645.1
38594.5
31963.1
13545.9
10073.3
4481.9
-1096.7
-12111.6
-18789.5
-21491.2
-21960.8
-22164.1
-28033.9
-29580.9
-46088.4
-47841.9
%
6.9%
3.6%
2.9%
2.7%
2.3%
1.0%
0.7%
0.3%
-0.1%
-0.9%
-1.3%
-1.5%
-1.6%
-1.6%
-2.0%
-2.7%
-3.3%
-3.4%
" bold = ±3.0% and beyond, italics =*from ±2.0% up to ±3.0% of grand mean; normal = within ±2.0% of grand mean
-------
Values within ±2.0% of the grand mean appear in normal font; values from ±2.0% up to ±3.0% of
the grand mean are shown in italics; and values that differ from the grand mean by ±3.0% or more
are shown in bold.
Subdividing the results into such categories allows the following observations to be made on the
probes tested:
• Within ±2.0% of grand mean: Both the manual and baseline Autoprobes, when operated
in yaw-nulled mode, were always within ±2.0% of the grand mean.
• Within ±3.0% of grand mean: The DAT probe, the manual Type S probe yaw-nulled, and
baseline Autoprobes straight-up were always within ±3.0% of the grand mean.
• ±3.0% and beyond of grand mean: All other probes (Type S straight-up, Prandtl, manual
Autoprobe straight-up, modified Kiel, French, spherical) had excursions beyond ±3.0% of
the grand mean at one or more sites. Of these probes, several showed consistent patterns
relative to the grand mean. That is, the manual Type S straight-up, Prandtl, manual
Autoprobes straight-up, and modified Kiel were always higher than the grand mean. At the
two sites where the flow was not near-axial (Lake Hubbard and Homer City), the spherical
and French probes were lower than the grand mean.
Although the variability of the Prandtl probe at the two gas sites where it was tested was the lowest
of any of the manual probes (coefficients of variation of 0.63% and 1.27% as shown in Section 3.3,
below), mean flow readings at both sites were higher than the central tendency. The Prandtl had the
highest difference from the central tendency (4.3%) at the near-axial site (DeCordova) and a less
severe excursion (2.2%) at the moderate-yaw/moderate-pitch site (Lake Hubbard). Because the
Prandtl probe is used as the standard to calibrate other probes in wind tunnels with axial flow, the
high difference at the near-axial site was unexpected. On the other hand, its more central behavior
at Lake Hubbard suggests that the flow reading taken with the probe may not be adversely affected
by moderate yaw and pitch. Note that the relative flow values determined by the Prandtl probe are
similar to those found for the manual Type S probe operated in the yaw-null mode.
The central tendency analysis using the grand mean provides an indication of how close the mean
volumetric flow obtained by a specific type of probe is to the grand mean for all methods. In
addition, line plots with various confidence intervals around the mean flow values for each run
corroborate the observations made in the grand mean analysis. Appendix F presents the plots.
3.2 WITHIN-METHOD ANALYSIS: ANALYSIS OF VARIANCE OF MATRIX B DATA
In addition to examining the relationships among the various methods, the importance of specific
variance components in the measurement of volumetric flow can be evaluated. The experimental
design enables the relative variability contributed to the measurement of volumetric flow due to
selection of test teams and the selection of different copies of a specific probe to be evaluated.
Assessment of these variance components is an important consideration in determining whether a
particular new method or procedure is appropriate for inclusion in a revised test method for
volumetric flow. Data generated from Matrix B provide the information for this analysis.
3-17
-------
3.2.1 ANOVA on Probe Copies
An ANOVA model with the groupings "probe copy" and "test team" was fit to the Matrix B data
(see Appendix E for a description of the model). The model was fit independently to data from each
type of probe (e.g., DAT, modified Kiel). In general, four copies of each probe type were analyzed.
The results are presented in Table 3-13. The columns labeled "ANOVA" indicate whether probe
copy is a significant factor. A "yes" in this column indicates that the choice among the four available
copies of the same probe type made a detectable difference in the flow value obtained. A "no" in
this column indicates that the choice of a particular copy had no detectable impact on the measured
volumetric flow.
Table 3-13. Results of ANOVA and Multiple Comparison Tests to Detect
Differences Among Copies of the Same Probe Type in the
Matrix B Data Set
Probe
DAT
Spherical A
Spherical B
Modified Kiel
Type S Yaw-nulled
Type S Straight-up
Prandtl
French
DeCordova
ANOVA
Yes
_b
-
No
No
Yes
No
No
Means Test*
M
_b
-
-
-
M
-
-
Lake Hubbard
ANOVA
No
No
-
-
No
No
-
-
Means Test1
-
-
-
-
-
-
-
-
Homer City
ANOVA
Yes
No
Yes
Yes
No
No
-
No
Means Test*
M
-
O (SPH-8)
M
-
-
-
-
* M=Multiple groups of probes identified, no individual outlier probes. O (Probe ID) = An individual outlier
probe identified, whose ID is shown in parentheses.
b Due to damage of two spherical probes, the Matrix B model could not be run; see Section 2.2.1.
In those cases where a "yes" appears in the ANOVA column, a multiple comparison means test
(Duncan's) was performed. The results are displayed in the column labeled "Means Test." If one
probe copy was found to yield significantly different results from all other copies, the letter "O"
(indicating "one copy") and the identifier code of the distinctive copy appears in this column. If the
multiple comparison test indicates that more than one copy yielded significantly different results, the
letter "M" appears in the "Means Test" column.
Table 3-13 shows that significant differences among probe copies most often occurred for the DAT
probe. The DAT probe copies were shown to have a significant effect at two of the three test sites.
However, as indicated by the letter "M" in the "Means Test" columns for these two sites, no
individual probe gave significantly different results from the other three DAT probes. It should be
recalled that whereas each test team supplied its own DAT probe, all other probes, including the
spherical probes, were manufactured at the same time and supplied by EPA. Multiple probe copy
effects (indicated the letter "M" in the "Means Test" columns) were also detected at one site for the
Type S and Kiel probes. In addition, at Homer City, one of the spherical probes (SPH-8) was
significantly different from the other spherical probe copies. No engineering or mechanical reasons
3-18
-------
were identified to explain the differences between measurements by this copy of the spherical probe
relative to measurements by the other copies of this type probe.
3.2.2 ANOVA on Test Teams
Table 3-14 shows that the test team effect was most often significant for the spherical probes. In all
three sets of Matrix B runs performed on the spherical probes at Lake Hubbard and Homer City, the
choice of test team significantly influenced the volumetric flow obtained. For the Spherical A
probes, one test team's volumetric flow values were significantly different from the others at both
Lake Hubbard and Homer City. For the Spherical B probes, the difference among teams was still
significant, but no single test team was significantly different from the rest (see Section 2.2.1 for
clarification of the definitions of the Spherical A and B probes).
Table 3-14. Results of ANOVA and Multiple Comparison Tests to Detect
Differences Among Test Teams in the Matrix B Data Set
Probe
DAT
Spherical A
Spherical B
Modified Kiel
Type S Yaw-nulled
S Straight-up
Prandtl
French
DeCordova
ANOVA
No
_b
-
No
No
No
Yes
No
MeansTesf
-
_b
-
-
-
-
M
-
Lake Hubbard
ANOVA
No
Yes
-
-
Yes
Yes
-
-
MeansTesf
-
O(Teaml)
-
-
M
M
-
-
Homer City
ANOVA
Yes
Yes
Yes
No
No
No
-
No
MeansTest"
M
O (Team 1)
M
-
-
-
-
-
a M=Multiple groups of teams identified, no individual outlier teams. O (Team ID) = An individual team outlier
identified, whose ID is shown in parentheses.
Due to damage of two spherical probes, the Matrix B model could not be run.
Results obtained using the French and modified Kiel probes were not significantly influenced by the
test team. The test team effect for the Prandtl probe was significant at the gas-fired site with near-
axial flow (DeCordova), for the Type S probe straight-up and yaw-nulled at the gas-fired site with
moderate yaw and pitch (Lake Hubbard), and for the DAT probe at the coal-fired site (Homer City)
with high yaw components of flow.
These results indicate that the choice of test team has no consistent effect across sites with the
possible exception of the spherical probes. Lack of a significant test-team effect for these methods
indicates that the equipment was operated in a consistent manner, which resulted in data with small
between test-team variance. In addition, except for the Spherical A probes at Lake Hubbard and
Homer City, no one test team produced results that were significantly different from the other three
teams. This finding indicates no problem, in general, with any specific probe, piece of equipment,
or test team procedure.
3-19
-------
The test-team effect, however, is likely to be more of a factor in routine test team performance of the
methods than was evident in these field tests. Despite efforts to recruit a representative set of testers
for this study, the test teams participating in an EPA-sponsored study of this magnitude, inevitably,
represent some of the most qualified and experienced testers in the field. In addition, the field tests
were deliberately performed under very controlled conditions with fewer time constraints than are
typically faced by testers when performing routine emissions testing. All these factors would tend
to minimize the differences among test team results. The differences in results among testers at large
could be expected to be considerably greater than those experienced in this study.
3.3 ANALYSIS OF METHOD VARIABILITY
Table 3-15 presents a statistical analysis of the variability in each method using the Matrix A data.
For each method, the coefficient of variation (CV, defined as the standard deviation divided by the
mean times 100) of the Matrix A data is shown. Because the data were obtained over four to eight
runs, the calculated CVs represent the temporal sampling variation associated with each probe. (This
variation should not be confused with the variance components analysis presented in Section 3.5,
where multiple sources of uncertainty are explicitly differentiated in the analysis.) A small CV
indicates that the probe provided consistent measurements over the course of Matrix A. For
example, a CV of 2% is equivalent to stating that the 95% confidence interval of the mean spans a
range of approximately 8%, i.e., 4% on either side of the mean.
A graphical display of the variability of the Matrix B data can be found in Appendix G.
Using both Table 3-15 and Appendix G, the following conclusions can be drawn:
• The variances associated with the Autoprobes and the engineering methods are consistently
small (CV < 2.3%) at all sites.
• At the two sites where they were tested, the Prandtl probes displayed the lowest variability
of all the manual methods (CVs of 0.63% at DeCordova and 1.27% at Lake Hubbard).
• The original spherical probes (Spherical A) at Lake Hubbard (high and low load) have the
highest CVs (19.70% and 7.39%, respectively).
• Relatively high CVs are associated with the modified Kiel probe (Lake Hubbard: 5.19%
and 4.04%); French probe (Homer City: 4.18%, Lake Hubbard high load: 3.28%); DAT
probe (Lake Hubbard: 3.02% and 3.14%); and Type S yaw-nulled probe (Lake Hubbard
high load: 3.23% and Homer City: 3.18%).
3.4 SUMMARY OF CENTRAL TENDENCY AND VARIABILITY ANALYSES
Figure 3-7 assembles the central tendency and variability analyses for the probe/mode combinations
tested in the field study. The figure allows a visual examination in two dimensions of the statistical
properties of the tested methods. The following conclusions can be drawn from this figure:
• The manual Autoprobe 16-point yaw-nulled and baseline Autoprobe 48-point straight-up
achieved the best (i.e., lowest) central tendency (always within ± 1 % of the grand mean) and
variability results (CVs always below 1%).
3-20
-------
Table 3-15. Method Variability Analysis for Matrix A Runs
Probe/Method
DAT
Type S Straight-up
Type S Yaw-nulled
Modified Kiel
Spherical A
Spherical B
Prandtl
French
Manual Autoprobe 16-point Straight-up
Manual Autoprobe 16-point Yaw-nulled
Baseline Autoprobes 16-point Straight-up
Baseline Autoprobes 16-point Yaw-nulled
Baseline Autoprobes 48-point Straight-up
Baseline Autoprobes 48-point Yaw-nulled
Baseline Autoprobes 12-point Straight-up
Baseline Autoprobes 12-point Yaw-nulled
O2 F-Factor
CO2 F-Factor
MMBtu
ASMEPTC4.1
BTCE
CV (%)"
DeCordova
2.66
1.88
2.61
1.39
0.90
NA
0.63
1.62
0.34
0.10
0.27
0.32
0.24
0.32
NA
NA
0.48
1.25
0.26
1.12
0.57
Lake Hubbard
High Load
3.02
2.31
3.23
5.19
19.70
NA
1.27
3.28
0.43
0.25
0.37
0.42
0.21
0.26
NA
NA
0.67
0.69
0.38
0.41
0.30
Low Load
3.14
1.04
1.76
4.04
7.39
NA
NA
NA
NA
NA
0.91
0.63
0.31
0.28
NA
NA
0.59
0.76
0.64
0.43
2.22
Homer City
2.08
1.58
3.18
1.24
0.93
1.42
NA
4.18
0.90
0.93
0.98
0.99
NA
NA
1.23
0.95
1.76
1.90
1.95
2.02
0.98
CV = coefficient of variation, (standard deviation / mean) x 100.
Among the remaining methods, the baseline Autoprobes yaw-nulled (12-, 16-, and 48-point)
always achieved the best (i.e., lowest) central tendency results (within ±2% of the grand
mean) and variability results (CVs below 1%).
The DAT, Type S yaw-nulled, and baseline Autoprobe 12-point and 16-point straight-up
had comparable central tendency results (always within ±3% of the grand mean), but the
variability of the baseline Autoprobes 12-point and 16-point straight-up was lower (CVs
consistently below 2%) than the DAT and Type S yaw-nulled (for which neither CV was
consistently below 3%).
3-21
-------
zz-c
•%£ Mopq A"j}U9}sisuoo jou '-9-1 'qSiq SJSM JBIJJ sjjnssj Ajiji
pus Abuspus} re.nu30 qjoq psq ssqojd y IBOusqdg pire 'prjj pgyipoui 'qousij
pire
rejis S
sj|ns9j Xouspusj {BJJU90 qSiq psq ssqoad ppUBij sqj pue dn-jqSrejjs §
'sunj
11V ••s»ins34 A>i|iqBUBA pue
<— Better Maximum Coefficient of Variance Worse —>
fO jw O>
§88
— 1
o
*
CO
•o
7
ical Probes 1-4 are off
&
i
o
9
g
i
3
ID
3
O
f s
i?
1
to
.00%
Maximum Percent
Is
» o
1 *
o
G>
u
Q.
i S
a 0
s
(D
o
> i i i i i i
• 2
i*
z
^
§ £ co
Z CD
2
M _ O
1* '
S* i
? -n
> * • i
' S-
O)
I *
-
• »
y &
* (D
?> . P 0
" 1 1
I f i
1* " §
-------
• The Spherical B probes had better central tendency (always within ±2% of the grand mean)
and variability results (CVs always below 2%) than the Spherical A (SPH 1-4) probes.
(Note that the Spherical A probe values are off the scales and so are not shown in the
figure.) Because both Spherical A and B probes, however, were available for testing only
at the last field test site (Homer City), these findings should not be considered conclusive.
Table 3-16 summarizes the results of the central tendency and variability analyses.
Table 3-16. Summary of Central Tendency and Variability
Analyses: Probe Achievement Levels"
Always Within
±1%
±2%
±3%
Beyond ±3%
Central Tendency
(% of Grand Mean)
BAP 48S
Man AP N
BAP 48N
BAP 16N
BAP 12N
Spherical B
BAP 16S
BAP 12S
Type S Yaw-nulled
DAT
French
Man AP S
Prandtl
Type S Straight-up
Modified Kiel
Spherical A
Variability (CV in %)
BAP 48S
BAP 16S
Man AP S
BAP 48N
BAP 16N
BAP 12N
Man AP N
BAP 12S
Prandtl
Spherical B
Type S Straight-up
Type S Yaw-nulled
French
Modified Kiel
DAT
Spherical A
a Results are based on four sets of Matrix A runs (at DeCordova, Lake Hubbard high load,
Lake Hubbard low load, and Homer City) except as follows: The Spherical B and Autoprobe
12-point results are based on one set of Matrix A runs (at Homer City), Prandtl on two (at
DeCordova and Lake Hubbard high load), and Autoprobe 48-point on three (at all except
Homer City).
3.5 UNCERTAINTY IN ENGINEERING METHODS AND PROBE MEASUREMENTS
A comparison of the expected uncertainty in the engineering and probe methods is presented in this
section. This analysis is an important consideration for addressing the field study's collateral goal
of seeking improvements that might reduce the reported disparities between the heat input values
derived from in-stack and engineering methods. Knowing the uncertainties associated with in-stack
and engineering methods reveals when the flow values from each method are sufficiently close that
they fall within each other's overlapping error bands.
3-23
-------
The analysis in this section is similar in intent to the uncertainty analysis presented in Section 2.3
and the variability analysis presented in Section 3.3. In this section, however, the uncertainty
analysis was performed in a manner (i.e., establishing sampling distributions for input parameters
and performing repeated trials) that enables the expected uncertainty in the volumetric flow values
generated by an engineering method to be directly compared with the expected uncertainty in a
probe's measurements on matched runs. This direct comparison enabled the relative uncertainty in
the methods at both gas-fired and coal-fired plants to be evaluated. The evaluation was limited to
the MMBtu engineering method and the Type S probe straight-up measurement. The results of the
error analysis using these approaches are considered to be representative of the expected uncertainty
in most of the engineering and probe-specific techniques. In addition, the MMBtu and Type S probe
straight-up are used extensively in many field applications. The analysis was performed at the
DeCordova and Homer City test sites. Monte Carlo techniques are used as a basis for the error
analysis.
As noted earlier in Section 2.3, error analysis, like that presented here, provides a measure of the
variability (i.e., uncertainty) inherent in a method, not its accuracy. Error analysis reveals the
variation that can be expected in a method's results based on the variation in its input parameters,
irrespective of whether the method is a good or poor indicator of the true flow value.
Monte Carlo simulation is a widely used technique for estimating the prediction error in values
generated by complicated models or equations. For each uncertain parameter in an equation, a
sampling distribution is generated. The sampling distribution represents the uncertainty in the
measurement of each parameter. For this analysis, the sampling distributions were assumed to be
normal. A distribution, including the sufficient statistics of the distribution (i.e., the sample mean
and variance), is assigned to each uncertain parameter in the model. The Monte Carlo procedure first
draws a random sample from each of the parameter distributions. The model is run for these values,
and the resulting volumetric flow value is recorded. This procedure is repeated many times (in this
analysis, 10,000 times) resulting in a distribution of the predicted values. From this predictive
distribution, the error in volumetric flow can be calculated (e.g., the variance of the predictive
distribution can be determined).
Because of the complexity of the MMBtu method, the first-order error analysis for this
method presented in Section 2.3 used only first-level equations. In contrast, the Monte Carlo
technique uses all equations in performing its analyses, thereby providing a full characterization of
the measurement error. Because the Monte Carlo technique is much easier to implement for
complicated models than first-order error analysis, it is particularly well suited for analysis of
approaches like the MMBtu method.
The equations for the MMBtu and Type S straight-up method are presented in Appendix H.
Table H-l in Appendix H presents the inputs to the MMBtu method for each test site. In addition,
the source of information used to calculate the sampling mean and variance are presented, along with
the CV for each parameter. Information used to generate the sampling distributions include the raw
data collected during the field studies, information concerning the allowable error as specified in 40
CFR Part 60 for specific methods, and expert judgement. For methods specified in 40 CFR Part 60,
the total error allowable was calculated as specified in the method.
3-24
-------
Table H-2 in Appendix H includes the same information for the parameters of the Type S probe
straight-up measurements. The sources of information used to derive the sampling variances include
allowable error specified in 40 CFR Part 60, expert judgement, and sensitivity testing of the equation
used to calculate volumetric flow.
Table 3-17 presents the results of the Monte Carlo analysis. The mean and standard deviation of the
predictive distribution generated by the Monte Carlo method is presented for each method at the
DeCordova and Homer City test sites. The CV is also presented.
Table 3-17. Monte Carlo Error Analysis Results on Volumetric Flow (wscfm)
Method
MMBtu
Type S Straight-up
DeCordova
Mean
1,539,302
1,612,854
s.d.
17,875
28,847
CV (%)
1.2
1.8
Homer City
Mean
1,430,653
1,452,558
s.d.
74,724
27,770
CV (%)
3.1
1.9
Examination of the table shows that the expected error in volumetric flow for the Type S probe
ranges between a CV of 1.8% and 1.9% across the test sites. The expected error in the MMBtu
method ranges, however, from a CV of 1.2% to 3.1%.
These results can be used to address the question of whether differences in flow measured by the
Type S probe straight-up and flow calculated by the MMBtu method are statistically significant.
Using the larger CV for each site, differences of less than 3.6% (1.8% times 2) between the two
measured values were within two standard deviations of each other and so were not statistically
distinguishable at the gas-fired site. Differences of less than 6.2% (3.1% times 2) were not
statistically distinguishable at the coal-fired site.
3-25
-------
SECTION 4
SUPPLEMENTAL DATA ANALYSIS
The analyses presented in this section were conducted in response to recommendations provided to
EPA by a panel of peer reviewers. This section also presents results of comparisons among various
pressure measurement devices. Section 4.1 presents a refinement of the central tendency analysis.
Section 4.2 presents a comparison of probe measurements to an engineering baseline. Section 4.3
presents a sensitivity analysis of the effect of excluding certain data points that appeared implausible
to peer reviewers, even though no documented problems could be found to justify their exclusion
based on the criteria established for identifying outliers. Section 4.4 presents a comparison of
pressure data collected by manual and electronic devices at two wind tunnels and several field test
sites.
4.1 REFINEMENT OF THE CENTRAL TENDENCY ANALYSIS
The concerns expressed by the peer reviewers regarding the Round 1 central tendency analysis
presented in Section 3 are summarized below.
Calculating the central tendency, or the grand mean, of the volumetric flow values using the
measurements obtained by every probe type used in the study may not be appropriate for the
following reasons:
• As shown in Section 3, some probe types used in the study (the modified Kiel and French,
for example) produced uniformly high or highly variable results. Values obtained by such
probes should not be included in the calculation of the grand mean.
• Some probe types and methods were over-represented in the Round 1 grand mean
calculations. For example, there were four Type S probes (manual Type S, manual
Autoprobe Type S, baseline Autoprobes 16-point, and baseline Autoprobes 48-point), each
of which was operated in both straight-up and yaw-nulled modes, and five engineering
methods (O2 F-factor, CO2 F-factor, ASME PTC 4.1, MMBtu, and BTCE). Because probes
or methods of the same type tend to produce similar flow measurements, a grand mean
value that makes use of all probe types and methods will be overly influenced by the results
of the over-represented probe types or methods, and therefore may not be a good estimate
of the true flow in the stack.
• Accounting for the yaw component of flow generally results in more accurate
measurements. Probes that do not account for yaw angles generally give higher readings
than those that do and should not be included in calculations of the grand mean.
The decay of flue gas velocity near the stack wall is a widely accepted phenomenon and is observed
for the wall-effects runs in this study. Because this decay is not adequately captured by a standard
Method 1 sampling traverse, the volumetric flow values used in the Round 1 grand mean
calculations do not account for wall effects and are therefore high. Adjusting the flow values
obtained by all probe types for wall effects should result in a more accurate grand mean.
Based on these concerns and suggestions and the results from the Round 1 analysis presented in
Section 3, a second central tendency analysis was performed, as described below. The intent of the
Round 2 analysis was to implement peer reviewers' recommendations that the grand mean be
4-1
-------
calculated only from the methods they considered to be most technically credible and that the results
be evaluated to see how close these methods come to both the grand mean and each other.
The refined grand mean was calculated using flow measurements from the following four probe
types and two engineering methods: manual Type S yaw-nulled, DAT, spherical (Run 3 at Lake
Hubbard was excluded because it seemed implausible to several peer reviewers), baseline
Autoprobes 16-point yaw-nulled, MMBtu, and BTCE. For DeCordova and Lake Hubbard, the
Autoprobes 16-point yaw-nulled, DAT probes, MMBtu, and BTCE each contributed eight flow
values to the grand mean, while the Type S yaw-nulled and spherical probes each contributed four
flow values. For Homer City, each probe type and method contributed four values to the grand
mean, except for the spherical probes, which contributed eight.
Flow measurements obtained by the four probe types were adjusted to account for wall effects prior
to calculation of the grand mean. The wall effects adjusted volumetric flow values were calculated
as follows:
Fad] = Funad] x (1 + A/100) Eq. 4-1
where:
Fadj = volumetric flow adjusted for wall effects;
Funadj = volumetric flow unadjusted for wall effects (from Method 1 points); and
A = wall effects adjustment factor.
The wall effects adjustment factors, A, were derived using data collected during wall-effects tests
at each of the three test sites. They are expressed in percent and bear a negative sign, for example,
-1.57% for the DAT probe at DeCordova. The factors for each probe type are presented in Table
11-1 in Appendix II-1.
Although not included in the Round 2 grand mean calculations, the Type S straight-up and Prandtl,
which are the current allowable Method 2 probe types, are included in the tables and graphs
presented in this section for comparison purposes.
Of the five engineering methods performed during the field tests (see Section 3), the MMBtu method
was included in the Round 2 analysis, because the peer review panel indicated that it was the most
widely accepted and commonly used of the four stoichiometry-based methods (O2 F-factor, CO2 F-
factor, MMBtu, and ASME PTC 4.1). The BTCE method was included because it is based on
conservation of energy rather than stoichiometry.
The results of the Round 2 central tendency analysis are summarized in Table 4-1. Following the
convention used earlier in Table 3-12 and explained in Section 3.1.4 in this report, Table 4-1 uses
a bi-directional ranking of methods, where the column labeled "rank" gives the highest positive rank
to the method with the greatest positive deviation from the grand mean and the lowest negative rank
to the method with the greatest negative deviation from the grand mean.
4-2
-------
Table 4-1. Summary Statistics for Round 2 Central Tendency Analysis (Matrix A)
Probe/Method
Vol Flow Grand
Mean (wscfm)
Comparison to Grand Mean
Mean Diff.
s.d.
Percent Diff.
DeCordova
Type S Straight-lip"
Prandtl11
Type S Yaw-nulled
DAT
Spherical
BL AP 16-point Yaw-nulled
MMBtu
BTCE
1,542,922.4
38,813.9
54,428.1
28,099.1
-3,854.3
27,815.6
-3,568.3
-9,517.0
-7,540.9
29,722.4
9,986.7
41,003.3
40,959.9
14,073.4
4,984.1
3,956.0
8,787.0
2.52%
3.53%
1.82%
-0.25%
1.80%
-0.23%
-0.62%
-0.49%
Lake Hubbard (High Load, Without Spherical Run 3)
Type S Straight-up"
Prandtl1
Type S Yaw-nulled
DAT
Spherical
BL AP 16-point Yaw-nulled
MMBtu
BTCE
1,073,469.6
61,044.1
22,307.9
18,759.4
-11,181.6
-37,086.3
-2,139.4
18,372.2
-523.5
26,192.1
13,863.3
35,249.5
32,108.5
13,984.5
4,534.3
4,106.0
3,182.6
5.69%
2.08%
1.75%
-1.04%
-3.45%
-0.20%
1.71%
-0.05%
Rank
3
4
2
-2
1
-1
-4
-3
4
3
2
-3
-4
-2
1
-1
Lake Hubbard (Low Load)
Type S Straight-up"
Prandtl1
Type S Yaw-nulled
DAT
Spherical
BL AP 16-point Yaw-nulled
MMBtu
BTCE
693,929.4
28,814.1
21,520.4
-8,648.6
-65,256.1
4,213.1
29,914.1
18,293.1
7,549.9
12,596.3
21,546.9
46,432.3
4,364.4
4,623.7
15,811.3
4.15%
3.10%
-1.29%
-9.40%
0.61%
4.31%
2.64%
4
3
-1
-2
1
5
2
Homer City
Type S Straight-up"
Type S Yaw-nulled
DAT
Spherical A
Spherical B
BL AP 16-point Yaw-nulled
MMBtu
BTCE
1,379,855.1
105,168.4
38,838.4
663.3
-13,069.1
-11,542.6
-7,360.1
12,630.6
-13,047.1
23,523.2
45,170.5
28,651.4
12,662.5
19,446.4
13,583.7
27,188.9
13,396.1
7.62%
2.81%
0.05%
-0.95%
-0.84%
-0.53%
0.92%
-0.95%
4
3
1
-4
-2
-1
2
-3
Not included in the grand mean calculation.
4-3
-------
The following observations can be made from examination of Table 4-1:
• The Round 2 analysis grand mean for each site was smaller than the Round 1 analysis grand
mean [0.84%, 1.49%, 2.58%, and 2.46% smaller for DeCordova, Lake Hubbard (high load),
Lake Hubbard (low load), and Homer City, respectively].
• At all sites, except Lake Hubbard at low load, all the methods included in the grand mean
produced flow values that were close to the grand mean. At DeCordova, the flow values
from all methods included in the grand mean were themselves within 2% of the grand
mean. At Lake Hubbard (high load) and Homer City, only one method that was included
in the grand mean produced values that were more than 2% from the grand mean—the
spherical probes (3.45%) at Lake Hubbard (high load) and the Type S yaw-nulled probes
(2.81%) at Homer City. In contrast, under the low load conditions at Lake Hubbard, four
of the six methods included in the grand mean were themselves more than 2% from the
grand mean.
• Of the in-stack methods, the Autoprobes 16-point yaw-nulled produced flow values that
were always within 0.61% of the grand mean, and the DAT probes produced values that
were always within 1.29% of the grand mean.
• The flow values for the Type S yaw-nulled probe were higher than the grand mean at all
three sites, ranging from 1.75% at Lake Hubbard (high load) to 3.10% at Lake Hubbard
(low load).
4.2 Comparison to Engineering Baseline
Some peer reviewers suggested that, at the gas-fired facilities, a mass balance calculation would yield
the most accurate representation of flue-gas flow rates. These reviewers therefore recommended that
the data analysis include comparisons of the in-stack probe measurements to an engineering baseline,
rather than solely to the baseline Autoprobes 16-point straight-up, as presented in Section 3. Such
a comparison had already appeared in Appendix A of the individual site data reports for the
DeCordova, Lake Hubbard, and Homer City field tests, where Series 13 and 14 tables and figures
used the BTCE and MMBtu methods, respectively, as baselines. Further comparisons are presented
here in response to the reviewers' recommendations.
Table 4-2 compares flow values for the methods analyzed in Section 4.1 to the values obtained by
the MMBtu method using Matrix A data only. The MMBtu method was selected as a baseline for
this comparison because the peer review panel thought it to be generally accepted and commonly
used by industry.
As in Table 4-1, Table 4-2 compares flow values for the methods selected for Round 2 analyses after
the velocity decline near the stack wall has been taken into account. It should be noted that the Lake
Hubbard high-load data exclude spherical Run 3, as requested by several peer reviewers. Table 4-2
presents the mean difference, standard deviation, and percent difference between each method and
the average MMBtu value at each site.
4-4
-------
Table 4-2. Summary Statistics for Round 2 Analysis Using MMBtu Method as the
Baseline (Matrix A)
Probe/Method
Average MMBtu
Vol. Flow (wscfm)
Comparison to MMBtu
Mean Diff.
s.d.
DeCordova
Type S Straight-Up
Prandtl
Type S Yaw-Nulled
DAT
Spherical
BL AP 16-point Yaw-Nulled
MMBtu
BTCE
1,533,405.4
48,330.9
63,945.1
37,616.1
5,662.8
37,332.6
5,948.8
0.0
1,976.1
29,722.4
9,986.7
41,003.3
40,959.9
14,073.4
4,984.1
3,956.0
8,787.0
Percent Diff.
3.15%
4.17%
2.45%
0.37%
2.43%
0.39%
0.00%
0.13%
Lake Hubbard (High Load, Without Spherical Run 3)
Type S Straight-Up
Prandtl
Type S Yaw-Nulled
DAT
Spherical
BL AP 16-point Yaw-Nulled
MMBtu
BTCE
1,091,841.9
42,671.9
3,935.6
387.1
-29,553.9
-55,458.5
-20,511.6
0.0
-18,895.8
26,192.1
13,863.3
35,249.5
32,108.5
13,984.5
4,534.3
4,106.0
3,182.6
3.91%
0.36%
0.04%
-2.71%
-5.08%
-1.88%
0.00%
-1.73%
Lake Hubbard (Low Load)
Type S Straight-Up
Prandtl
Type S Yaw-Nulled
DAT
Spherical
BL AP 16-point Yaw-Nulled
MMBtu
BTCE
723,843.5
-1,100.0
NA
-8,393.8
-38,598.8
-95,170.3
-25,701.0
0.0
-11,621.0
7,549.9
NA
12,596.3
21,546.9
46,432.3
4,364.4
4,623.7
15,811.3
-0.15%
NA
-1.16%
-5.33%
-13.15%
-3.55%
0.00%
-1.61%
Homer City
Type S Straight-Up
Type S Yaw-Nulled
DAT
Spherical A
Spherical B
BL AP 16-point Yaw-Nulled
MMBtu
BTCE
1,392,485.8
92,537.8
26,207.8
-11,967.4
-25,699.8
-24,173.3
-19,990.8
0.0
-25,677.8
23,523.2
45,170.5
28,651.4
12,662.5
19,446.4
13,583.7
27,188.9
13,396.1
6.65%
1.88%
-0.86%
-1.85%
-1.74%
-1.44%
0.00%
-1.84%
4-5
-------
With respect to the in-stack methods, the primary observation that can be made from Table 4-2 is
the closeness to the MMBtu values of the measurements by the Autoprobes 16-point yaw-nulled
(0.39%, -1.88%, -3.55%, and -1.44%), the Type S yaw-nulled (2.45%, 0.04%, -1.16%, and 1.88%),
and the DAT probes(0.37%,-2.71%,-5.33%, and-0.86%). This observation suggests that provision
for the use of these two types of probes might address one of the collateral goals of the field study,
that is, to reduce the disparity that industry had reported between in-stack and combustion-based
calculations of heat rate and flow.
4.3 SENSITIVITY ANALYSIS ON LAKE HUBBARD SPHERICAL PROBE DATA
As described in Section 2, the criterion used in this field study for identifying values as outliers and
excluding them from the analytical data set was the existence of documented procedural or
equipment problems. Any value that differed substantially from other values in the same or proximal
runs and any value that exceeded the 1.5-IQR statistical screening criterion were flagged and
investigated, but only those for which procedural or equipment problems could be documented were
excluded from analysis as outliers.
A number of peer reviewers questioned the plausibility of several Lake Hubbard spherical probe
values that remained in the analytical data set after application of the criterion described above. In
particular, they questioned the values obtained by probe SPH-4 on Run 3, by probe SPH-2 on Run
22, and by probe SPH-4 on Run 24. (Individual probe values are presented by run in Tables 3A
through 3E in Appendix A of the Lake Hubbard Site Data Report.)
Based on these comments, the three identified values were re-examined. No new evidence of
procedural or equipment problems was found. However, whereas the values obtained on Runs 22
and 24 were consistent with proximal values obtained by other probes, the value obtained on Run
3 clearly was not. Of the three questioned values, the Run 3 value was also the only one that was
implausible in view of the unit load at the time of the test.
A sensitivity analysis was therefore performed to evaluate the impact on the analysis presented in
Section 3 of removing the Run 3 value for probe SPH-4 from the analytical data set. The results are
presented in Tables 4-3 through 4-8. The most noticeable impact of removal of the SPH-4, Run 3
value was observed in the following results:
• In the central tendency analysis, removal of the SPH-4, Run 3 value brought the spherical
probe closer to the grand mean, from -12.3% to -4.0%. By changing the grand mean itself,
the removal also brought the modified Kiel, Type S straight-up and yaw-nulled, and Prandtl
probes closer to the central tendency of the data and caused the DAT and French to depart
somewhat further from the central tendency (Table 4-3).
• The CV for the spherical probes in the Lake Hubbard high-load runs dropped from 19.70%
to 1.35%. The CV in the low-load runs at Lake Hubbard (7.39%) was unaffected by the
removal of the Run 3 data (Table 4-4).
4-6
-------
Table 4-3. Sensitivity Analysis on Round 1 Central Tendency Analysis for Matrix A
Runs—Lake Hubbard (High Load) (Based on Table 3-12)"
With SPH-4, Run 3
Grand Mean = 1,089,755.58 wscfm
Rank
11
10
9
8
1
6
5
4
3
2
1
-1
-2
-3
-4
-5
-6
-7
Method
Modified Kiel
Type S Straight-up
Type S Yaw-nulled
Prandtl
Manual AP Straight-up
ASMEPTC4.1
BL AP 16-point Straight-
up
CO2 F-factor
Manual AP Yaw-nulled
MMBtu
BL AP 48-point
Straight-up
BLAP 16-point
Yaw-nulled
DAT
BL AP 48-point
Yaw-nulled
O2 F-factor
BTCE
French
Spherical A
Difference
wscfm
79,070.4
65,906.9
23,968.2
23,726.2
14,430.3
13,568.9
12,650.9
6,434.3
3,227.7
2,086.3
1,988.5
-1,226.4
-10,633.5
-10,777.2
-15,322.1
-16,809.5
-22,242.1
-133,505.
6
%
7.3%
6.0%
2.2%
2.2%
1.3%
1.2%
1.2%
0.6%
0.3%
0.2%
0.2%
-0.1%
-1.0%
-1.0%
-1.4%
-1.5%
-2.0%
-12.3
%
Without SPH-4, Run 3
Grand Mean = 1,093,563.40 wscfm
Rank
8
7
6
5
4
3
2
1
-1
-2
-3
-4
-5
-6
-7
-8
-9
-10
Method
Modified Kiel
Type S Straight-up
Type S Yaw-nulled
Prandtl
Manual AP Straight-up
ASMEPTC4.1
BLAP 16-point
Straight-up
CO2 F-factor
Manual AP Yaw-nulled
MMBtu
BL AP 48-point
Straight-up
BLAP 16-point
Yaw-nulled
DAT
BL AP 48-point
Yaw-nulled
O2 F-factor
BTCE
French
Spherical A
Difference
wscfm
75,262.6
62,099.1
20,160.4
19,918.4
10,622.5
9,761.1
8,843.1
2,626.5
-580.1
-1,721.5
-1,819.3
-5,034.2
-14,441.3
-14,585.0
-19,129.9
-20,617.3
-26,049.9
-43,316.4
%
6.9%
5.7%
1.8%
1.8%
1.0%
0.9%
0.8%
0.2%
-0.1%
-0.2%
-0.2%
-0.5%
-1.3%
-1.3%
-1.7%
-1.9%
-2.4%
-4.0%
bold = ±3.0% of grand mean and beyond, italics =from ±2.0% up to ±3.0% of grand mean; normal = within ±2.0% of
grand mean.
4-7
-------
Table 4-4. Sensitivity Analysis on Method Variability Analysis For Matrix A Runs
(Based on Table 3-15)
Method
DAT
Type S Straight-up
Type S Yaw-nulled
Modified Kiel
Spherical A
Spherical A w/o SPH-4, Run 3 at LH
Spherical B
Prandtl
French
Manual Autoprobe 16-point Straight-up
Manual Autoprobe 16-point
Yaw-nulled
Baseline Autoprobes 16-point
Straight-up
Baseline Autoprobes 16-point
Yaw-nulled
Baseline Autoprobes 48-point
Straight-up
Baseline Autoprobes 48-point
Yaw-nulled
Baseline Autoprobes 12-point
Straight-up
Baseline Autoprobes 12-point
Yaw-nulled
O2 F-Factor
CO2 F-Factor
MMBtu
ASMEPTC4.1
BTCE
CV (%)
DeCordova
2.66
1.88
2.61
1.39
0.90
NA
0.63
1.62
0.34
0.10
0.27
0.32
0.24
0.32
NA
NA
0.48
1.25
0.26
1.12
0.57
Lake Hubbard
High Load
3.02
2.31
3.23
5.19
19.70
1.35
NA
1.27
3.28
0.43
0.25
0.37
0.42
0.21
0.26
NA
NA
0.67
0.69
0.38
0.41
0.30
Low Load
3.14
1.04
1.76
4.04
7.39
NA
NA
NA
NA
NA
0.91
0.63
0.31
0.28
NA
NA
0.59
0.76
0.64
0.43
2.22
Homer City
2.08
1.58
3.18
1.24
0.93
1.42
NA
4.18
0.90
0.93
0.98
0.99
NA
NA
1.23
0.95
1.76
1.90
1.95
2.02
0.98
CV = coefficient of variation, (standard deviation / mean) * 100.
4-8
-------
The removal had limited, but discernible, impact on the following results:
• The mean percent difference of the spherical probes from the baseline Autoprobes 16-point
straight-up in Matrices A and B changed from -7.17% to -5.86% (Table 4-5).
Table 4-5. Sensitivity Analysis on Volumetric Flow Summary—Lake Hubbard (Based
on Table 3-7)
Probe/Method
DAT
Type S Straight-up
Type S Yaw-nulled
Modified Kiel
Spherical
Spherical w/o SPH-4 for Run 3
Prandtl
French
Manual Autoprobe 1 6-point
Straight-up
Manual Autoprobe 1 6-point
Yaw-nulled
Baseline Autoprobes 1 6-point
Straight-tip"
Baseline Autoprobes 1 6-point
Yaw-nulled
Baseline Autoprobes 48-point
Straight-up
Baseline Autoprobes 48-point
Yaw-nulled
O2 F-Factor
CO2 F-Factor
MMBtu
ASMEPTC4.1
BTCE
Number
of Tests
29
25
25
8
25
24
4
4
4
4
25 ,
25
24
24
25
25
25
25
25
% Difference from Baseline"
Mean
-2.98%
3.65%
1.72%
3.11%
-7.17%
-5.86%
0.93%
-3.24%
0.09%
-0.93%
0.00%
-1.08%
-0.99%
-2.06%
-1.75%
-0.02%
-0.50%
0.49%
-2.31%
Minimum
-13.44%
0.98%
-3.70%
-4.24%
-38.73%
-17.81%
-0.30%
-7.37%
-0.71%
-1.58%
0.00%
-2.93%
-1.48%
-3.13%
-3.60%
-1.78%
-1.99%
-0.81%
-3.31%
Maximum
0.54%
7.92%
4.93%
12.04%
-1.22%
-1.22%
3.12%
-0.16%
0.97%
-0.29%
0.00%
0.47%
-0.18%
-0.88%
0.78%
2.69%
1.21%
2.37%
1.60%
s.d.
3.10%
1.83%
2.16%
4.71%
7.84%
4.37%
1.52%
3.16%
0.75%
0.52%
0.00%
0.66%
0.36%
0.51%
1.12%
0.96%
0.78%
0.76%
0.99%
Baseline is baseline Autoprobes 16-pohlt straight-up.
4-9
-------
• The analysis of variance changed slightly. For the original ANOVA Variant 1, the
engineering methods were not distinguishable from the 1-D and 2-D probes, but all were
different from the 3-D probes. Without Run 3, the engineering methods and 3-D probes
could not be distinguished, but both differed from the 1-D and 2-D probes (Table 4-6).
Table 4-6. Sensitivity Analysis on Duncan's Multiple Range Test—Lake Hubbard
(High Load) (Based on Table 3-11)
Variant
Variant 1"
Variant 2 b
With SPH-4, Run 3
(ID, 2D, Eng) * 3D
(2Dm, lDm, 1DJ * 3Dm
Without SPH-4, Run 3
(ID, 2D) * (Eng, F, 3D)
2Dm, lDm, ID. * 3Dm
(2Dm, 1DJ * (2Da, F, 3DJ
2DmMDa,2D1,F,3Dm
Variant 1 Classes
ID = Type S straight-up, baseline Autoprobes straight-up, manual Autoprobe straight-up, Prandtl
2D = Type S yaw-nulled, baseline Autoprobes yaw-nulled, manual Autoprobe yaw-nulled, modified Kiel
3D = DAT, spherical
F = French
Eng = Engineering methods (O2 F-factor, CO2 F-factor, MMBtu, ASME PTC 4.1, BTCE)
b Variant 2 Classes
lDm = manual 1-D methods (Type S straight-up and Prandtl)
2Dm = manual 2-D methods (Type S yaw-nulled and modified Kiel)
3Dm = 3-D methods (DAT and spherical)
lDa = Autoprobes 1-D methods (baseline Autoprobes straight-up)
2Da = Autoprobes 2-D methods (baseline Autoprobes yaw-nulled)
F = French
The following results were largely unchanged by the removal of the SPH-4, Run 3 value:
• overall range in all flow values (Table 4-7), and
• in-stack average vs. engineering average flow values (Table 4-8).
Table 4-7. Sensitivity Analysis on Overall Range of Flow Measurements—Lake
Hubbard (High Load) (Based on Table 3-1)
Max (wscfrn)
Min (wscfrn)
Difference
With SPH-4, Run 3
Avg. Value
1,157,629
1,036,472
121,157 (11.02%)b
Method
Modified Kiel1
Spherical
Without SPH-4, Run 3
Avg. Value
1,157,629
1,054,582
103,047 (9.77%)b
Method
Modified Kiel"
Spherical
The modified Kiel probe produced the highest measurement of flow when measurements taken only under high-load
operation were included. When low-load measurements were also included, the Type S probe straight-up gave the
highest average measured flow. See Tables 3-6 through 3-8.
Ratio of difference relative to average of all in-stack and engineering methods.
4-10
-------
Table 4-8. Sensitivity Analysis on In-stack and Engineering Methods
Comparison—Lake Hubbard (High Load) (Based on Table 3-4)"
Comparison
In-stack Avg. vs. Engineering Avg.
S Straight-up Avg. vs. Engineering Avg.
With SPH-4 for Run 3
0.26% (20)
5.44% (8)
Without SPH-4 for Run 3
0.46% (20)
5.44% (8)
a Number of runs analyzed is shown in parentheses.
4.4 COMPARISON OF MANUAL AND ELECTRONIC PRESSURE MEASURING DEVICES
Some peer reviewers expressed the opinion that manual pressure devices (i.e., the manometer and
magnehelic gauges) provide consistently higher pressure readings than electronic transducers. To
investigate this assertion, a series of comparisons was made of pressure readings taken with manual
and electronic devices during test programs conducted as part of this study and by others. Data from
the following tests were used: an EPA-sponsored swirl tunnel test conducted by Fossil Energy
Research Corp. (FERCo); an EPRI-sponsored field test at the Coal Creek and Columbia power plants
conducted by FERCo; an EPA-sponsored field test conducted by Radian (now known as the Energy
Research Group); and a NIST wind tunnel test during which the probes used in this study were
calibrated. In both the wind tunnel tests and the field tests, manual and electronic pressure devices
were used in parallel to measure pressures.
Analysis of the five data sets produced the following findings:
• For different types of probes and different velocity conditions, no one type of pressure
device consistently produced higher or lower measurements.
• Even where statistically significant differences were found between the manual and
electronic devices, the differences in pressure readings were, from a practical standpoint,
generally small.
• With Type S probes, manual devices tended to read higher than electronic transducers in
the field tests. In the wind tunnel tests, the manual devices tended to provide lower pressure
readings.
• The percent differences between the manual and the electronic devices were generally
smaller on average and less variable in the wind tunnel tests than in the field tests.
• With electronic transducers, very little difference was found between taking a straight
average of the individual differential pressure measurements (Approach 1) and squaring the
average of the square roots of the individual pressure measurements (Approach 2). For the
data examined, the average differential pressure using Approach 1 was always higher than
the average differential pressure using Approach 2, but never by more than 0.44%.
A detailed description of the devices used, the type of data collected, and the analyses performed,
and a fuller discussion of the findings appear in Appendix 12.
4-11
-------
SECTION 5
WALL EFFECTS STUDY
A key element of EPA's evaluation of potential improvements to the Agency's current test methods
for volumetric flow was the study of velocity decay near the stack wall, referred to as the "wall
effect." The primary objective of the wall effects study was to investigate the impact of velocity
decay close to the stack wall on volumetric flow. Collateral objectives were to (1) evaluate a
procedure for adjusting for velocity decay close to the stack wall using actual field data and (2)
obtain information on practical aspects of implementing the procedure under conditions typically
encountered at electric utility stacks.
Wall effects tests were conducted at each of the three steel-stack field test sites during the day under
steady load conditions (Matrix C runs) using manual probes and Autoprobes, and at night under
nonsteady-state load conditions using only Autoprobes. Additional wall effects tests using only
Autoprobes were conducted at six sites with brick and mortar stacks in order to investigate the effect
on volumetric flow of velocity decay for rough stack surfaces. All of the over-night tests and some
of the day-time tests at these additional sites were conducted under nonsteady-state load conditions.
Because of the importance of flow stability in evaluating changes in velocity before and after
adjustment for velocity decay close to the stack wall, all wall effects data collected under nonsteady-
state conditions were examined for velocity stability by the procedure described in Exhibit J-l in
Appendix J before being used in the study. Data sets where the velocity at the Method 1 equal-area
traverse points closest to the wall in the original (baseline) traverse differed by more than 8% from
the velocity at the corresponding traverse point in the wall effects traverse were excluded from the
analyses reported in this section. In addition, due to design features of some probes and/or physical
constraints at some testing locations, measurements could not be made beginning at 1 in. from the
stack wall for some probes at some field test sites. Data sets that did not contain a complete set of
measurements beginning at 1 in. from the stack wall also were excluded from the analyses reported
in this section. Table J-l in Appendix J provides a complete summary of all wall effects data sets
collected in the study that met the velocity stability criterion.
5.1 WALL EFFECTS DATA COLLECTION PROCEDURE
In the wall effects tests, a standard 12-, 16-, or 20-point cross-stack traverse was first performed as
prescribed in Method 1. The velocity was measured at the centroid of each equal-area sector and the
average cross-stack velocity was calculated from the resulting data for these baseline traverses.
Next, a wall effects traverse was performed in which velocity measurements were taken at 1-in.
increments across the entire width of each of the four Method 1 equal-area sectors adjacent to the
stack wall, usually starting at 1 in. from the wall and moving toward the center of the stack. As
noted above, for some types of probes and for some sampling ports, obtaining measurements within
a few inches of the stack wall was not possible due to the size and shape of the probe, because the
sampling port nipple extended beyond the interior stack wall into the gas stream, or because the
annular space in the stack did not allow moving the probe to within 1 in. of the stack wall. In such
instances, the wall effects sampling procedure allows the first 1-in. incremented velocity
measurement to be made at a distance of greater than 1 in. from the stack wall. In calculating the
replacement velocity value (see following paragraph) in these cases, the procedure stipulates that the
first actual measurement made in a sampling port should be used for all sampling points closer to
5-1
-------
the stack wall where measurements cannot be made. The data in Table J-l in Appendix J include
some tests where measurements could not be made beginning at 1 in. from the stack wall in all ports.
For these tests, the replacement velocity values were calculated in accordance with this provision
of the procedure. However, only data sets where measurements could be made beginning at 1 in.
from the stack wall were used in the analyses reported in this section.
The wall effects traverse measurements were used to calculate a replacement velocity value for the
Method 1 traverse point closest to the stack wall for each of the four sampling ports. The
replacement velocity values were calculated as described in Section 5.2 below. It should be noted
that although measurements were taken across the entire width of each of the four Method 1 equal-
area sectors adjacent to the stack wall in the field study in order to gather information about the
decay in velocity close to the stack wall, the procedure followed in the study ("Method 2 Draft
Revisions: Wall Effects Velocity Procedure"7) could easily be adapted to provide for sampling as
few as two points. The calculational procedure described in Section 5.2 is much simpler when
implemented for only a few points.
The average cross-stack velocity was then recalculated after replacing the measured velocity values
for the four Method 1 traverse points closest to the stack wall with the calculated replacement
velocity values. The percent difference between the original average cross-stack velocity and the
average cross-stack velocity making use of replacement velocity values was then calculated. The
calculated percent difference indicates the effect on volumetric flow due to velocity decay close to
the stack wall.
5.2 CALCULATION OF REPLACEMENT VELOCITY VALUES USING WALL EFFECTS DATA
Replacement velocity values for 12- and 16-point traverses were calculated in accordance with the
procedure presented in Form 5-1A (or 5-IB), "Method 2 Draft Revisions: Wall Effects Velocity
Procedure." This procedure was adapted for calculating replacement velocity values for the 20-point
traverses that are reported in Table J-l in Appendix J. Exhibit J-2 in Appendix J provides an
example of a replacement velocity calculation for a 16-point traverse.
5.3 SUMMARY OF WALL EFFECTS DATA OBTAINED FOR THE STUDY
Wall effects tests were conducted at each of the three field test sites (DeCordova, Lake Hubbard, and
Homer City) using manual probes and Autoprobes. Because the three field test sites all have smooth
steel stacks, additional wall effects tests were conducted at the six sites with brick and mortar stacks
listed below in order to investigate the effect on volumetric flow of velocity decay for rough stack
surfaces. The additional wall effects tests at these sites were conducted using only Autoprobes and
were not observed by EPA or Cadmus personnel. While not subject to the same degree of on-site
scrutiny as the tests at DeCordova, Lake Hubbard, and Homer City, the tests at these sites were
conducted by the same test personnel, using the same or identical test equipment, following the same
test protocol as was used at the three primary field tests sites. For these reasons, it was decided to
include the data in the study.
See, for example, The Cadmus Group, Inc., 1997, "Flow Reference Method Testing and Analysis: Field Test Plan, DeCordova
Electric Station," EPA/430/R-97-024.
5-2
-------
Sites for Additional Wall Effects Tests
American Electric Power Co., Inc. • American Electric Power Co., Inc.
Columbus Southern Power Columbus Southern Power
Conesville Units 1/2, 3, and 5 Pic way Unit 9
General Public Utilities Corp.
Metropolitan Edison Co.
Titus Unit 1
Allegheny Power System, Inc.
West Penn Power Co.
Mitchell Unit 3
Table 5-1 shows key characteristics of the sites for the additional wall effects tests. See Table 1-2
for corresponding key characteristics of DeCordova, Lake Hubbard, and Homer City.
Table 5-1. Key Characteristics of Sites for Additional Wall Effects Tests
Station
Location
Stack No.
Stack Inside Diameter
Test Dates
Unit
Fuel
Boiler Type
Nameplate Capacity
Load Range During Tests8
Conesville
Conesville, OH
1/2
17.2ft
5/16/98
1
Bituminous Coal
B&W
1. OOOxl 06lb/hr
cyclone
148 MWe
79-84 MWe
2
Bituminous Coal
B&W
1. OOOxl 06lb/hr
cyclone
136 MWe
79-88 MWe
3
19.0ft.
3/25/98
3
Bituminous Coal
Riley Stoker
1.100xl06lb/hr
front-fired
162 MWe
75-77 MWe
5 f
34.0 ft.
1/15-16/98
5
Bituminous Coal
CE
3.131 xl06lb/hr
tangential-fired
444 MWe
Day: 379-433 MWe
Night: 218-433 MWe
Station
Location
Stack No.
Stack Inside Diameter
Test Dates
Unit
Fuel
Boiler Type
Nameplate Capacity
Load Range During Tests"
Picway
Lockbourne, OH
9
14.5ft
1/13/98
9
Bituminous Coal
Riley Stoker
0.900 x 106lb/hr
front-fired
106 MWe
82-84 MWe
Mitchell
Courtney, PA
3
22.0ft
12/17/97
3
Bituminous Coal
CE
1.978x 106lb/hr
tangential-fired
299 MWe
Day:
16-point: 112-1 17 MWe
12-point: 2 12-2 17 MWe
Night: 204-267 MWe
Titus
Reading, PA
1
13.9ft
2/4-5/98
1
Bituminous Coal
CE
0.570 xl06lb/hr
tangential-fired
75 MWe
Day: 66-80 MWe
Night: 55-80 MWe
Load range is across all tests and does not necessarily indicate the range for any single test. Refer to Exhibit J-l in
Appendix J for an explanation of the velocity stability criterion that applies for individual tests.
5-3
-------
The data from the wall effects tests at DeCordova, Lake Hubbard, Conesville 1/2, Picway, and
Mitchell that met the completeness and velocity stability criteria, discussed earlier, are reported in
this section. The following paragraphs provide summary descriptions of the wall effects tests
conducted at each site reported in this section.
DeCordova. Seventeen wall effects tests were conducted at DeCordova during the day, seven using
manual probes and ten using Autoprobes. Fourteen additional wall effects tests were conducted at
DeCordova at night using Autoprobes. The data from all of these tests were used in developing the
findings reported in this section, except for one test where modified Kiel probes were used and one
test where spherical and DAT probes were used together (with DAT probes used in two ports and
spherical probes in the other two). The data from the modified Kiel probe test were excluded
because measurements could not be taken beginning at 1 in. from the stack wall. The data from the
spherical/DAT probe test were excluded from analysis in this section because the results could not
be ascribed to a single type of probe. The spherical probe data (obtained at two of the four ports
during the test) were used in deriving the wall effects adjustment factor shown in Table 4-1 in
Section 4.
Lake Hubbard. At Lake Hubbard, three wall effects tests were conducted during the day: one
using DAT probes and one each using Autoprobes operated in the straight-up and yaw-nulled modes.
Eighteen additional wall effects tests were conducted at Lake Hubbard at night using Autoprobes.
The data from all of these tests were used in developing the findings reported in this section.
Conesville Unit 1 and 2 Stack. Twelve wall effects tests were conducted at the Conesville Unit 1
and 2 shared stack during the day using Autoprobes. The data from all of these tests were used in
developing the findings reported in this section.
Picway. Twelve wall effects tests were conducted at Picway during the day using Autoprobes. The
data from all of these tests were used in developing the findings reported in this section.
Mitchell. Sixteen wall effects tests were conducted at Mitchell during the day using Autoprobes.
Six tests (three 12-point straight-up and three 12-point yaw-nulled) did not meet the velocity stability
criterion described in Appendix J, and results for these tests were not included in Table J-l in
Appendix J or used in developing the findings reported in this section.
Other Test Sites. No data sets from wall effects tests conducted at Homer City, Conesville Unit 3,
Conesville Unit 5, or Titus contained a complete set of measurements beginning at 1 in. from the
wall for all ports. Thus, the data from these sites were excluded from the analyses reported in this
section.
Exhibits J-3 through J-l 1 in Appendix J provide additional information on the wall effects tests
conducted at each site, including descriptions of the nature and limitations of the data collected at
the sites not reported in this section. The Series 7 tables in Appendix A of the DeCordova, Lake
Hubbard, and Homer City site data reports contain the data listed below for all the day-time tests at
these sites. Tables Ol through O3 in Appendix O of the site data reports contain analogous data for
the over-night tests at DeCordova, Lake Hubbard, and Homer City. Tables J-2(A-C) through J-7(A-
C) in Appendix J of this report contain these data for tests at the other sites. These tables contain
the following data: (1) velocity values for wall effects traverse points; (2) original (unadjusted) and
wall effects-adjusted velocity values for Method 1 points; and (3) percent differences between the
original (unadjusted) and wall effects-adjusted average cross-stack velocity values.
5-4
-------
Wall effects calculation forms for all tests (day-time and over-night) at DeCordova, Lake Hubbard,
and Homer City are located in Appendix P of the DeCordova and Lake Hubbard site data reports and
Appendix R of the Homer City site data report. The wall effects calculation forms for tests at the
other sites are located in Attachments A through F of Appendix J of this report.
5.4 ANALYSIS OF WALL EFFECTS DATA
Table 5-2 summarizes the wall effects data for tests where measurements could be made beginning
at 1 in. from the stack wall and that met the previously described velocity stability criterion. Unless
otherwise indicated, the results in Table 5-2 are based on 16-point baseline traverses. In columns
4 through 12 of Table 5-2, results are tabulated by probe type and, for manual Type S probes and
Autoprobes, by operational mode (straight-up and yaw-nulled). The Autoprobes 16-point data for
DeCordova and Lake Hubbard shown in columns 6 and 7 of Table 5-2 are the combined results for
Matrix C tests conducted during the day and the additional wall effects tests conducted at night. The
day-time and over-night 16-point Autoprobes results are shown separately in Table J-8 in Appendix
J. The last two columns in Table 5-2 show results aggregated across probe types and operational
modes for 16-point and 12-point tests.
Table 5-2. Summary of Percent Difference Between Original and Wall Effects-
Adjusted Average Velocity
Site
TU
DeCordova
TU
Lake
Hubbard
AEP
Conesville
Unit 1/2
AEP
Picway
Allegheny
Mitchell
Averages*
Stack
Mal'l
Steel
Steel
Brick
and
mortar
Brick
and
mortar
Brick
and
mortar
Steel
Brick
and
mortar
Steel and
brick
and
mortar
Statistics
No. Tests
Average
s.d.
No. Tests
Average
s.d.
No. Tests
Average
sd.
No. Tests
Average
s.d.
No Tests
Average
sd.
No. Sites
Average
sd.
No. Sites
Average
sd.
No. Sites
Average
sd.
Manual Type S
Straight-
Dp
1
-1.83%
1
-1.83%
;•:•:.;•'••• •- :•">
'•':- • !
••::.:•:'•;••••':
1
-1.83%
Yaw-
Nulled
1
-1.93%
1
-1.93%
1
-1.93%
Baseline Autoprobes
16 Point
Straight-
UP
13
-1.50%
0.52%
10
-1.41%
0.93%
3
-1.94%
0 24%
3
-1.92%
0.18%
4
-1.75%
1 83%
2
-1.46%
006%
3
-1.87%
0.10%
5
-1 70%
0 24%
Yaw-
Nulled
11
-1.51%
0.59%
10
-1.27%
0.97%
3
-1.92%
0.17%
3
-1.95%
0 23%
4
-1.69%
1 66%
2
-1.39%
0.17%
3
-1 85%
0.14%
5
-1 67%
0.29%
12 Point
Straight-
HP
3
-4 23%
0 12%
3
-2.37%
039%
1
-201%
3
-2.87%
1 19%
3
-2.87%
1.19%
Yaw-
Nulled
•.::--:;, .-;.•
3
-4.28%
0.25%
3
-2.53%
040%
1
-2.03%
3
-2.95%
1 18%
3
-2 95%
1.18%
DAT
1
-1 57%
1
-1.56%
•.-.•.•:.•,;:.
••.. . :. ' :
2
-1.57%
0 01%
2
-1 57%
0 0!%
Prandtl
1
-1 59%
:. .;..::•. •:,,
1
-1.59%
1
-1.59%
French
1
-1.13%
/.::,...,;..;.
1
-1 13%
::;: .. . .
V!".x'v:--"
1
-1.13%
Aggregate
Results'
16 Point
7
-1.58%
0.26%
3
-1.41%
0.15%
2
-1.93%
0 01%
2
-1.94%
0.02%
2
-1.72%
0 04%
2
-1 50%
0.12%
3
-1.86%
0 12%
5
-1 72%
0 23%
12 Point
! - !'-."'
. ,::.. .:..
2
-4 26%
0.04%
2
-2 45%
0. 1 1%
2
-2 02%
0 01%
3
-291%
1 19%
3
-2.91%
1.19%
Based on equal weighting of each probe type/mode of operation for site-designated rows and equal weighting of each site for stack
material-designated rows.
Based on equal weighting of average results for each site.
5-5
-------
The first five three-row blocks in Table 5-2 present results by site. The averages and standard
deviations shown in columns 4 through 12 of these blocks are based on equal weighting of each test.
The averages and standard deviations shown in the last two columns of these blocks are based on
equal weighting of average results for each probe type and operational mode. For example, the
average aggregate 16-point result for Lake Hubbard (-1.41%) is based on equal weighting of the
average of the ten 16-point Autoprobes straight-up tests (-1.41%), the average of the ten 16-point
Autoprobes yaw-nulled tests (-1.27%), and the single DAT test (-1.56%). This approach ensures that
results aggregated across probe types and modes of operation are not artificially weighted toward
those probe types/modes of operation for which a greater number of tests were performed.
The last three three-row blocks in Table 5-2 present results aggregated according to type of stack,
with the first of these three-row blocks presenting results for steel stacks, the second presenting
results for brick and mortar stacks, and the third presenting results for all stack types. Results in all
columns of these last three three-row blocks are based on equal weighting of average results for each
site. For example, the average aggregate 16-point result for brick and mortar stacks (-1.86%) is
based on equal weighting of the average result for each of three sites: Conesville Unit 1/2 (-1.93%),
Picway (-1.94%), and Mitchell (-1.72%). This approach ensures that results aggregated according
to stack type are not artificially weighted toward those sites where a greater number of tests were
performed.
5.4.1 General Findings
Weighting the results for each probe type/mode of operation and site equally, as described above,
the average percent difference in velocity across all probe types and sites for tests using 16-point
Method 1 traverses was -1.72% (s.d. 0.23%, n = 5 sites). With the exception of one 16-point test
where there was no (0%) change in velocity, three 16-point tests where the wall effects-adjusted
velocity ranged from 0.15% to 1.13% higher than the unadjusted velocity, and four 16-point tests
where the percent difference values (-4.33% and -3.87% for brick and mortar stacks and -2.82% and
-2.82% for steel stacks) were substantially larger than the next largest percent difference values (see
Section 5.4.10), the remaining 62 16-point wall effects tests analyzed for this report resulted in a
difference in velocity ranging from -0.61% to -2.36%. For 12-point traverses, the average percent
difference in velocity across all probe types and sites was -2.91% (s.d. 1.19%, n = 3 sites), with
results across 14 tests ranging from -2.00% to -4.49%.
5.4.2 Point-to-Point Percent Change in Velocity
Figures 5-1 through 5-5 illustrate the percent change in velocity between adjacent traverse points
(averaged across all ports) for all probes in the Matrix C (day-time) wall effects tests at DeCordova
and Lake Hubbard (Figures 5-1 and 5-2) and for Autoprobes tests at Picway, Mitchell, and
Conesville Unit 1/2 (Figures 5-3 through 5-5). The figures indicate that at distances farther than 5
to 7 in. from the stack wall, the average percent change in velocity between adjacent points either
varied positively and negatively from point to point (Lake Hubbard, Picway, and Mitchell) or
changed very little from point to point (DeCordova and Conesville Unit 1/2 stack). Beginning at a
distance of 5 to 7 in. from the stack wall, the average percent change in velocity between adjacent
points became significantly and consistently more negative when moving from point to point toward
the stack wall, indicating an acceleration in velocity decay in this region.
5-6
-------
Percent Change in Velocity Between Adjacent Wall Effects Points
Change Measured from Higher Numbered Point to Indicated Point
20%
10%
0%
0)
c
(0
§ -20%
«> -30%
Q.
-40%
-50%
2 3 4 5 67 8 9 10 11 12 13 14 15 16 17 18
Distance of Wall Effects Point from Stack Wall, in Inches
— Median change | Range of change
Box = 25th to 75th percentile
Figure 5-1. Point-to-point percent change in velocity across all
probes and ports at DeCordova.
Percent Change in Velocity Between Adjacent Wall Effects Points
Change Measured from Higher Numbered Point to Indicated Point
20%
10% -
a>
o>
n>
O
1 2 34 5 6 1 8 9 10 11 12 13 14 15 16 17
Distance of Wall Effects Point from Stack Wall, in Inches
— Median change | Range of change
Box = 25th to 75th percentile
Figure 5-2. Point-to-point percent change in velocity across all
probes and ports at Lake Hubbard.
5-7
-------
Percent Change in Velocity Between Adjacent Wall Effects Points
Change Measured from Higher Numbered Point to Indicated Point
20%
10%
0) 0%
.2 -10%
o
c -20%
-30%
-40%
-50%
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Distance of Wall Effects Point from Stack Wall, in Inches
— Median change | Range of change
Box = 25th to 75th percentile
Figure 5-3. Point-to-point percent change in velocity across all
ports at Picway.
Percent Change in Velocity Between Adjacent Wall Effects Points
Change Measured from Higher Numbered Point to Indicated Point
20%
10% -
-50%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Distance of Wall Effects Point from Stack Wall, in Inches
— Median change | Range of change
Box = 25th to 75th percentile
Figure 5-4. Point-to-point percent change in velocity across all
ports at Mitchell.
5-8
-------
Percent Change in Velocity Between Adjacent Wall Effects Points
Change Measured from Higher Numbered Point to Indicated Point
o>
20%
10%
0%
.c -10%
o
I -2°%
2
0) -30%
-40%
-50%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Distance of Wall Effects Point from Stack Wall, In Inches
— Median change | Range of change
Box = 25th to 75th percentile
Figure 5-5. Point-to-point percent change in velocity across all
ports at Conesville Unit 1/2 stack.
5.4.3 Findings With Respect to Stack Material
Based on the 16-point aggregate results summarized in Table 5-2, the average percent difference in
velocity due to wall effects was -1.50% (s.d. 0.12%, n = 2 sites) for smooth stacks (DeCordova and
Lake Hubbard) and -1.86% (s.d. 0.12%, n = 3 sites) for rough stacks (Conesville Unit 1/2, Picway,
and Mitchell). The observed larger average percent difference for rough stacks is qualitatively
consistent with theoretical models of the effects of stack walls on velocity.
Further analysis was performed to determine whether the average percent differences in velocity due
to wall effects in smooth and rough stacks were statistically significantly different from each other.
For this analysis, the aggregated data used to obtain the results in Table 5-2 were not suitable. The
results appearing in Table 5-2 were aggregated to avoid the possibly confounding effects on the
overall average that could arise from (1) different numbers of tests using each probe type/mode of
operation at any given site and (2) different numbers of tests from site to site. However, the
aggregating process reduces the variability associated with the overall average, as reflected in the
small standard deviations noted above and shown in the last two columns in Table 5-2. Retaining
such variability is critical in evaluating whether two means are statistically significantly different.
For this reason, in this phase of the analysis, the average and standard deviation for the 16-point steel
stack data and brick and mortar stack data were calculated using the unaggregated percent difference
values (n = 50 tests for steel stacks and n = 20 tests for brick and mortar stacks). The upper and
lower 95% confidence limits around the means also were calculated, as follows:
UCL = PD + cc
Eq. 5-1
5-9
-------
LCL = PD - cc
Eq. 5-2
where:
CC = '
0.025, n-\
Eq. 5-3
UCL
LCL
PD
cc
^0.025, n-l
n
upper 95% confidence limit (%),
lower 95% confidence limit (%),
average percent difference for steel stacks or brick and mortar stacks (%),
95% confidence coefficient (%),
t-distribution critical value for a probability of 0.025 and n-1 degrees of
freedom,
standard deviation of percent difference values for steel stacks or brick and
mortar stacks (%), and
number of tests used in developing the summary statistics.
The results of the analysis for the unaggregated data are shown in Table 5-3. For purposes of
comparison, Table 5-3 also shows the results for data aggregated by probe type/mode of operation
at each site and then averaged across sites, as presented in the next-to-last column of the sixth and
seventh three-row blocks in Table 5-2. Figure 5-6 displays the average percent difference in velocity
due to wall effects for steel stacks and for brick and mortar stacks, along with the corresponding
upper and lower 95% confidence limits around these mean values, for the unaggregated data.
Table 5-3. Summary Statistics on Percent Difference in Velocity for
Steel Stacks and for Brick and Mortar Stacks Based on
Unaggregated Data and Data Aggregated by Site
Stack Type
Data Aggregation
No. of Tests Averaged (n)
Average
Standard Deviation
95% Confidence Coefficient
95% Confidence Interval
Steel
None"
50
-1.45%
0.70%
0.20%
-1.25% to
-1.65%
Probe/site"
2
-1.50%
0.12%
1.08%
-0.42% to
-2.58%
Brick and Mortar
None"
20
-1.85%
1.00%
0.47%
-1.38% to
-2.32%
Probe/site"
3
-1.86%
0.12%
0.30%
-1.56% to
-2.16%
Based on equal weighting of each test.
Based on equal weighting of the averages for each site, where site averages are based on equal
weighting of the average results for each probe type/mode of operation at the site; e.g., for steel
stacks, the values in Table 5-3 are the same as the values in the 16-point aggregate results column
of the sixth three-row block of rows (steel stack) in Table 5-2.
5-10
-------
-1.0%
-1.2%
§ -1.4%
8j
o
I •1-8%
o
£ -2.0%
-2.2%
-2.4%
^
-
_
-
-
-
-
i
i
_
Steel Brick and Mortar
—
Stack Material
Average j 95% Confidence interval
Figure 5-6. Average and 95% confidence intervals of percent
difference in velocity, by stack material, based on
equal weighting of each test.
Table 5-3 shows that while data aggregation has little effect on the average percent differences for
steel stacks and brick and mortar stacks, the standard deviations for the unaggregated data are larger
than those for the aggregated data.
Figure 5-6 suggests that the wall effect for brick and mortar stack walls is likely to be greater than
that for steel walls, which is qualitatively consistent with theoretical models of wall effects for
different types of wall material and attendant roughness. However, the magnitude of the differences
is small: 0.40% for the unaggregated data and 0.36% for the aggregated data. In addition, using a
t-test to determine whether the average percent difference in velocity for steel stacks is statistically
significantly different from the average percent difference for brick and mortar stacks revealed that
the difference was not statistically significant at the 95% confidence level. This is reflected in the
overlap of the confidence interval lines shown in Figure 5-6.
5.4.4 Site-to-Site Comparisons
Across all sites, the average percent difference in velocity due to wall effects for any site was always
within 0.53% of that of every other site, based on 16-point aggregate results. There was only a
0.17% difference in the average percent difference in velocity due to wall effects between the two
sites with steel stacks. For the three sites with brick and mortar stacks, the difference in average
percent difference in velocity due to wall effects between any two sites was always less than or equal
to 0.22%.
5.4.5 Findings Relating to the Number of Points in the Original Traverse
The 12- and 16-point Autoprobes data collected at the Conesville Unit 1/2 stack, Pic way, and
Mitchell shown in Table 5-2 indicate that the percent difference between original and wall-effects
adjusted average velocity becomes larger as the number of points in the original traverse decreases.
5-11
-------
This finding is attributed primarily to the way wall effects adjustments were calculated in the study.
For the calculational procedure used in the study, as the number of points in the original traverse
increases, a smaller proportion of the traverse points are replaced with a wall effects-adjusted value.
That is, in a 12-point traverse, one third (four out of 12) of the baseline traverse points are replaced
by wall effects-adjusted values; whereas, in a 16-point traverse, one fourth (four out of 16) are
replaced. Thus, using the procedure employed in the study, the lower the number of baseline
traverse points, the greater is the impact of any velocity decay near the wall. As shown by the
aggregate 16- and 12-point percent difference values in the last two columns of the next-to-last three-
row block in Table 5-2, the average percent difference in velocity due to wall effects across the three
brick and mortar stack sites was-2.91% (s.d. 1.19%,n = 3 sites)for 12-point traverses and-1.86%
(s.d. 0.12%, n = 3 sites) for 16-point traverses.
Not only was the average percent difference in velocity due to wall effects larger for 12-point
traverses than for 16-point traverses, but the average velocities after adjustment for wall effects for
12-point traverses were lower, on average, than the average velocities after adjustment for wall
effects for 16-point traverses. This finding can be seen in Table 5-4, which shows the average
velocities for the 16- and 12-point Autoprobes tests at the Conesville Unit 1/2 stack, Picway, and
Mitchell. Table 5-4 presents the average unadjusted and wall effects-adjusted velocities for 16- and
12- traverses for tests in which both 16- and 12-point traverses were conducted.8 Across the three
sites, the average unadjusted velocities for 12-point original traverses were 0.79% higher, on
average, than the unadjusted velocities for 16-point traverses9; whereas, after adjustment for wall
effects, the average velocities for 12-point traverses were 0.40% lower than the average velocities
for 16-point traverses. It should be noted that the average wall effects-adjusted velocity for 12-point
traverses was lower than that for 16-point traverses due to the results for one of the three sites
(Conesville 1/2). For the other two sites, the wall-effects adjusted velocity for 12-point traverses was
essentially the same as that for 16-point traverses.
Table 5-4. Summary of Unadjusted and Wall Effects-Adj usted Velocities for 16- and 12-
Point Traverses
Site
Conesville 1/2
Picway
Mitchell
Average
Average Baseline Velocity (ft/sec)
16-Point
54.49
37.85
36.54
12-Point
55.02
38.06
36.85
% Difference'
0.97%
0.55%
0.85%
0.79%
Average Adjusted Velocity (ft/sec)
16-Point
53.43
37.11
36.05
12-Point
52.68
37.12
36.11
% Difference*
-1.40%
0.03%
0.17%
-0.40%
(12-point - 16-point)/16-point
5.4.6 Probe Type Comparisons
Findings regarding differences in wall effects attributable to probe type are somewhat tenuous
because data from different types of probes were not obtained at a sufficient number of sites to allow
Only one of four 12-point traverses conducted at Mitchell met the previously described velocity stability criterion. Thus, data
from only the one of four corresponding 16-point tests at Mitchell shown in Table 5-2 are reflected in Table 5-3.
This is qualitatively consistent with the relationship between 12- and 16-point Autoprobes data observed throughout the flow
study in which 12-point traverse data are almost always higher than 16-point data.
5-12
-------
definitive findings to be derived. While data were obtained at all sites for Autoprobes, data were
obtained at only one site (DeCordova) for Type S, Prandtl, and French probes and at only two sites
(DeCordova and Lake Hubbard) for DAT probes. Table II-1 in Appendix II shows wall effects
adjustment factors (i.e., percent differences) by probe type based on those Matrix C tests at
DeCordova and Lake Hubbard where measurements could be made beginning at 1 in. from the stack
wall in all ports. As can be seen from this table, Type S probes operated in the yaw-nulled mode
produced the largest wall effects adjustment factor, followed in decreasing order of wall effects
adjustment factor by Type S probes operated in the straight-up mode, Prandtl, DAT, and Autoprobes
operated in the yaw-nulled mode.10 Spherical probes produced the smallest wall effects adjustment
factor, which was based on results for two of four ports sampled in the test at DeCordova. 'It is
emphasized that these findings regarding wall effects adjustment factors and the relative order of
probe types with respect to producing wall effects adjustment factors are somewhat tenuous due to
the small number of sites where data were obtained for the Type S, DAT, Prandtl, French, and,
especially, the spherical probes.
5.4.7 Effect of Stack Gas Velocity
As shown in Figure 5-7, which is based on all tests involving 16-point original traverses, there is no
obvious relationship between the percent change in velocity due to wall effects and the average
velocity in the baseline traverse.
2%
1%
c
O)
1 -2%
O
§ -3%
-5%
C
*
* *
- ** I iV$ %
*
1 1 1 1 1
) 20 40 60 80 100 120
Average Velocity (fps) in Original Traverse
Figure 5-7. Percent change in velocity due to wall effects vs. average
velocity in the original traverse.
10
Table 11-1 does not include a wall effects adjustment factor for Autoprobes operated in the straight-up mode because this probe
type was not included in the analyses for which Table 11-1 was prepared. The Autoprobes operated in the straight-up mode
produced approximately the same wall effects adjustment factor as the DAT probes.
5-13
-------
5.4.8 Findings on Practical Aspects of Wall Effects Testing
Due to the size and shape of the probe head, it was not possible to obtain measurements using
modified Kiel probes at distances closer than 4 in. from the stack wall, which is the region where
most of the decay in velocity occurs. For this reason, modified Kiel probes are not suitable for use
in wall effects testing. (See Sections 6 and 7 for additional findings pertaining to equipment and
procedures that are relevant to wall effects testing.)
5.4.9 Wall Effects Calculations Using Minimum Number of Traverse Points
While as many 1 -in. incremented points as possible were sampled in the field tests in order to gather
information about the velocity decay close to the stack wall, a simpler implementation of the wall
effects procedure could involve sampling as few as two points per port: one 1 -in. incremented point
and d^. For comparison with the results obtained using all data, the calculational procedure was
carried out using only the velocity measurements made at the point located at 1 in. from the stack
wall and drem in each of the four Method 1 sectors adjacent to the stack wall. The percent differences
between average stack velocities before and after adjustment for wall effects, calculated both using
all data and using only two data points, are presented for each test in Tables J-9 (steel stacks) and
J-10 (brick and mortar stacks) in Appendix J. These tables list results for only those tests that met
the completeness and velocity stability criteria noted above. Table 5-5 provides summary statistics
for the data listed in Tables J-9 and J-10.
Table 5-5 shows that the percent differences calculated using only two data points tend to be smaller
in magnitude, or less negative, than their counterparts calculated using all data points. Across all
tests at all stacks, percent difference values calculated using all data points ranged from -4.33% to
+1.13%, with an average of-1.56% (s.d. 0.81%, n=70 tests), whereas the values calculated using
only two data points ranged from -1.79% to +2.41%, with an average of-0.72% (s.d. 0.66%, n=70
tests).
5.4.10 Maximum and Minimum Percent Differences
The maximum and minimum percent differences (decreases) in average velocity due to adjustment
for wall effects were -2.83% and +1.13%, respectively, for tests at steel stacks (see Table J-9 in
Appendix J), and -4.33% and +0.15%, respectively, for tests at brick and mortar stacks (see Table
J-10 in Appendix J). The positive percent differences clearly are anomalous. These anomalous
values may be due to a slight overall increase in the stack velocity between the times when
measurements were made in the baseline traverse and in the wall effects traverse (even though the
data met the previously discussed velocity stability criterion) or to imprecision (random fluctuations)
in the baseline and/or wall effects traverse measurements. The first hypothesis is supported by the
fact that the anomalous minimum value for steel stacks was obtained during overnight tests at Lake
Hubbard when the load was not as closely controlled as during the day-time tests. Similarly, the
anomalous minimum value for brick stacks was obtained during a test at Mitchell, where the load
may not have been controlled as closely as during the DeCordova and Lake Hubbard day-time tests.
The minimum negative percent differences in average velocity for steel and brick and mortar stacks
were -0.61 % (see Table J-9 in Appendix J) and -1.23% (see Table J-10 in Appendix J), respectively.
The two largest percent differences for tests at steel stacks (-2.82% and -2.82%) (see Table J-9 in
Appendix J) and for tests at brick and mortar stacks (-4.33% and -3.87%) (see Table J-10 in
Appendix J) also are suspicious in that they are significantly larger than the next largest percent
5-14
-------
difference values. These anomalous maximum values also may be due to imprecision in the baseline
and/or wall effects traverse measurements or to a slight overall decrease in the stack velocity
between the times when measurements were made in the baseline traverse and in the wall effects
traverse. This latter hypothesis is supported by the fact that the anomalous maximum values for steel
stacks were obtained during overnight tests at DeCordova when the load was not controlled as
closely as during the day-time tests. Similarly, the anomalous minimum values for brick stacks were
obtained during tests at Mitchell where anomalous minimum values also were obtained. Excluding
the possibly anomalous maximum and next-to-maximum values, the next largest percent differences
were -2.36% for tests at steel stacks (see Table J-9 in Appendix J) and -2.21% for tests at brick and
mortar stacks (see Table J-10 in Appendix J).
Table 5-5. Comparison of Percent Differences Between Original and Wall
Effects-Ad justed Average Velocities, Using All Data Points and
Using Only Two Data Points per Port
All Stacks
Smallest (Least Negative) % Difference
Smallest Negative % Difference
Largest (Most Negative) % Difference
Average (Excluding Non-negative % Differences)
Standard Deviation
Average (Using All % Differences)
Standard Deviation
% Difference Using
All Data Points
1.13%
-0.61%
-4.33%
-1.69%
0.62%
-1.56%
0.81%
% Difference Using
Two Data Points
2.41%
-0.08%
-1.79%
-0.87%
0.37%
-0.72%
0.66%
Steel Stacks
Smallest (Least Negative) % Difference
Smallest Negative % Difference
Largest (Most Negative) % Difference
Average (Excluding Non-negative % Differences)
Standard Deviation
Average (Using All % Differences)
Standard Deviation
% Difference Using
All Data
1.13%
-0.61%
-2.82%
-1.55%
0.23%
-1.45%
0.70%
% Difference Using
Two Data Points
0.20%
-0.08%
-1.71%
-0.74%
0.36%
-0.71%
0.39%
Brick and Mortar Stacks
Smallest (Least Negative) % Difference
Smallest Negative % Difference
Largest (Most Negative) % Difference
Average (Excluding Non-negative % Differences)
Standard Deviation
Average (Using All % Differences)
Standard Deviation
% Difference Using
All Data Points
0.15%
-1.23%
-4.33%
-2.06%
0.80%
-1.85%
1.00%
% Difference Using
Two Data Points
2.41%
-0.39%
-1.79%
-1.08%
0.37%
-0.75%
1.08%
5-15
-------
SECTION 6
FINDINGS
Drawing on the analyses presented in Sections 3 and 4, the findings for each of the tested probes and
related issues are presented below.
6.1 DAT PROBE
The DAT probe gave favorable results with respect to approaching the central tendency of the data
(within ±1.0% of the grand mean at DeCordova, Lake Hubbard high load, and Homer City and 2.3%
below the grand mean at Lake Hubbard low load). This finding is supported by the refined central
tendency analysis in Section 4 (DAT probes were always within ±1.3% of the grand mean across all
sites). At Lake Hubbard and Homer City (the sites with non-axial flow), the flow values measured
with the 3-D probes (DAT and spherical) statistically distinguishable from flow measured with the
1-D probes (Type S straight-up, French, and Prandtl), while they were not at DeCordova (the site
with axial flow).
The DAT probe's variability was typical of manual probes (the coefficient of variation ranged from
2.08% at Homer City to 3.14% for the low-load runs at Lake Hubbard), but was significantly higher
than that typical of automated probes. Conversely, the DAT probe had the largest range of ranks in
the rank-order analysis of volumetric flow, which suggests high run-to-run variability. Such
variability is also evident in the confidence interval plots of the DAT probe data. This variability
is not unexpected for a probe that measures three separate pressure drops for each velocity
determination. In addition, the differences in probe copies and test team operation described below
contribute to probe variability.
The ANOVA analyses of volumetric flow indicate that at two sites probe copy was a significant
factor in volumetric flow obtained by the DAT probes; that is, some variation in flow is attributable
to different copies of the probe. In addition, at one site the test team was a significant factor in flow
variation. The differences in results produced by different copies of the DAT probe are not
surprising because, unlike with all other probes, which were provided by Cadmus, each test team
supplied its own DAT probe.
6.2 PRANDTL PROBE
Although the coefficient of variation of the Prandtl probe at the two gas sites where it was tested was
the lowest of any of the manual probes (0.63% and 1.27%), mean flow readings at both sites were
higher than the central tendency. The Prandtl probe had the highest deviation from the central
tendency (4.3%) at the near-axial site (DeCordova) and a less severe excursion (2.2%) at the
moderate yaw/moderate pitch site (Lake Hubbard). This finding is consistent with the refined central
tendency analysis in Section 4 (3.53% from the central tendency at DeCordova and 2.08% at Lake
Hubbard). Because the Prandtl is used as a standard to calibrate other probes in wind tunnels with
axial flow, the high deviation at the near-axial site is unexpected. On the other hand, its more central
behavior at Lake Hubbard suggests that the flow reading taken with the probe may not be adversely
affected by moderate yaw and pitch. At Decordova, the only site where a Matrix B test was
performed on the Prandtl probe, a test team effect but no probe copy effect was detected. It is
interesting to note that the flow values determined by the Prandtl probe are similar to those found
by the manual Type S probe operated in the yaw-null mode.
6-1
-------
6.3 SPHERICAL PROBE
The four original spherical probes (damaged at DeCordova and subsequently repaired and re-
calibrated before Lake Hubbard) produced moderately high flow values at DeCordova and by far the
lowest flow values for both the high- and low-load tests at Lake Hubbard (12.3% and 10.6%,
respectively, below the grand mean). If Run 3 is excluded, as suggested by peer reviewers, the
average flow measurement for spherical probes for Lake Hubbard high-load tests is lower than the
grand mean by only 3.45%. At Lake Hubbard low-load, the percent deviation from the grand mean
was three times greater than that of any other method. On the other hand, at Homer City where both
the new set and old set of spherical probes were tested, the volumetric flow values were very close
to each other and much closer to the central tendency of the data (2.0% and 2.1% below the grand
mean). The rank order of the spherical probes was on the high end at DeCordova, lowest at Lake
Hubbard, and in the middle at Homer City. Consistent results were observed from the refined central
tendency analysis in Section 4, except that the deviation from the central tendency was smaller across
all sites.
At Lake Hubbard, the site with the highest traverse point-to-traverse point variation in flow velocity
and flow angle, the spherical probe's variability as reflected in the CV of the flow measurements was
19.70% at high load (nearly four times that of any other probe type) and 7.39% at low load (nearly
twice that of any other probe type). Again, if Run 3 is excluded, the coefficient of variation for the
high-load test drops significantly from 19.70% to 1.35%. At the other test sites, the variability of
flow values was comparable to those of other manual probes. An inspection of the confidence
interval plots in Appendix D reveals that the high variability at Lake Hubbard high load was due to
a dramatically lower flow value determined by one spherical probe operated by one test team during
a single run, whereas at Lake Hubbard low load it was due to swings in the spherical probe's flow
values from run to run. Inspection of the dispersion analysis graphs in Appendix G also indicates
greater swings and spread among the flow values obtained by different copies of the spherical probe
across all test sites, as compared to other probe types. Of all probes tested, test team effect was most
pronounced for the spherical probes. In all Matrix B (intraprobe comparison) runs at Lake Hubbard
and Homer City, the choice of test team was a significant factor in measured volumetric flow. In
addition, one of the new spherical probes was found to produce flow measurements that were
statistically different from the other three probes even though all four probes had the same design
and dimensions and were fabricated by the same manufacturer.
6.4 AUTOPROBES YAW-NULLED
Of all the tested probe types, the baseline Type S Autoprobe system operated in the yaw-nulled mode
with 16-point traverses was closest to the central tendency across all three sites and load levels, as
found in both the original and refined central tendency analyses (0.3%, -0.1 %, -0.4%, and -1.6% for
DeCordova, Lake Hubbard high load and low load, and Homer City, respectively, from the central
tendency in the original analysis, and -0.23%, -0.20%, 0.61%, and -0.53% for DeCordova, Lake
Hubbard high load and low load, and Homer City, respectively, from the central tendency in the
refined analysis). Results for 12- and 48-point traverses and the single manual Autoprobe were
comparable. The baseline 12-, 16-, and 48-point Autoprobes yaw-nulled and the manual Autoprobe
yaw-nulled displayed the lowest variability among all the methods tested, with coefficients of
variation consistently below 1%. For example, the coefficients of variation of the baseline
Autoprobes 16-point yaw-nulled were 0.32%, 0.42%, 0.63%, and 0.99% at DeCordova, Lake
Hubbard high load and low load, and Homer City, respectively. These central tendency and
6-2
-------
variability results are confirmed by visual inspection of the confidence interval plots in Appendix
D and dispersion plots in Appendix G.
6.5 AUTOPROBES STRAIGHT-UP
Straight-up operation of the Autoprobes consistently produced higher flow values than yaw-nulled
operation. As expected, the difference between the straight-up and yaw-nulled modes increased from
near-axial to moderate yaw/moderate pitch to high-yaw flow conditions. For example, the percent
differences between the yaw-nulled and straight-up 16-point baseline Autoprobes were 0.14%, 1.08%,
4.33% for DeCordova, Lake Hubbard, and Homer City, respectively. For 16-point traverses, the
deviation from the grand mean of straight-up operation increased from 0.3% at DeCordova, to 1.2%
and 0.9% for Lake Hubbard high load and low load, to 2.7% for Homer City. At the near-axial site
(DeCordova) the deviation from the grand mean for straight-up operation (for 16-point, 48-point, or
manual 16-point) was no more than 0.2% larger than the deviation from the grand mean for yaw-
nulled operation. At the other sites, differences between the straight-up and yaw-nulled values were
substantially larger (up to 1.4% at Lake Hubbard and 4.4% at Homer City). The low variability
observed for yaw-nulled operation was also found for straight-up operation. The coefficient of
variation of the manual and baseline Autoprobes 12-, 16- and 48-point straight-up never exceeded
1.23% at any site. The central tendency and variability results for the straight-up operation with 48
traverse points at the two gas-fired sites (i.e., within 0.3% of the central tendency with coefficients
of variation less than 0.31%) were comparable to the results discussed in the previous section for the
Autoprobes operated in yaw-nulled mode.
6.6 TYPE S PROBE YAW-NULLED
The Type S probe operated in yaw-nulled mode consistently produced flow results 2.2% to 2.9%
higher than the grand mean. The results of the refined central tendency analysis in Section 4 are
consistent with these results (1.75% to 3.10% above the central tendency). As stated in Section 6.2,
the percent difference from the grand mean and the average volumetric flow values for the Type S
probe yaw-nulled were very similar to those for the Prandtl probe at the two gas-fired sites, where
Prandtl probes were tested. At each site the average volumetric flow values measured by the two
probes differed by less than 1 %. The coefficients of variation for the Type S probe yaw-nulled across
all three sites ranged from 1.76% to 3.23%, which was typical for the manual probes tested. No
significant differences among probe copies were found, but at Lake Hubbard some test team effect
on flow measurement was detected. As with the Autoprobes, yaw-nulled operation consistently
produced lower flow values than straight-up operation. As expected, the difference between the two
modes increased from near-axial to moderate yaw/moderate pitch to high-yaw flow conditions, where
the differences between the two modes, derived in terms of the percent difference from the baseline
Autoprobes 16-point straight-up data, were 0.50%, 1.93%, 4.60% for DeCordova, Lake Hubbard, and
Homer City, respectively). These differences between the two modes are similar to those observed
for the Autoprobes.
6.7 TYPE S PROBE STRAIGHT-UP
The Type S probe operated in the straight-up mode has been the standard method for measuring
volumetric flow for over 20 years. The rank order analysis showed the median rank of its volumetric
flow values to be highest or second highest at each of the three field test sites. The Type S probe
straight-up had the highest positive difference (6.9%) from the central tendency at Homer City and
in low-load operation at Lake Hubbard (3.4%), second highest for high-load operation at Lake
6-3
-------
Hubbard (6.0%), and third highest at DeCordova (3.5%). In the refined central tendency analysis in
Section 4, the Type S probe straight-up had the highest positive deviation from the central tendency
at Lake Hubbard high load and low- load, and Homer City (5.69%, 4.15%, and 7.62%, respectively),
and the second highest at DeCordova (2.52%). At the same time, its measurements at the near-axial
site (DeCordova) were slightly lower than those of the Prandtl, which is the accepted standard pilot
used to calibrate other probes in wind tunnels with axial flow. The coefficient of variation of the
Type S straight-up measurements was 2.31% or less at all three sites. Some probe copy effect was
detected at DeCordova and some test team effect was detected at Lake Hubbard, but neither of these
effects was detected at the coal-fired site (Homer City).
6.8 FRENCH PROBE
At the moderate yaw/moderate pitch and high-yaw angle sites (Lake Hubbard and Homer City), the
French probe produced volumetric flow values that were lower than the central tendency (-2.0% and
-3.3%). At Homer City, the French probe produced the lowest measurements of all the tested in-stack
methods (3.3% below the grand mean). The French probe's coefficient of variation ranged from
1.62% at the gas-fired site with axial flow (DeCordova) to 4.18% at the coal-fired site, where it was
the most variable of all tested methods. Its variation in rank order relative to the central tendency,
from +6 at DeCordova to -6 at Lake Hubbard and -8 at Homer City, suggests a strong dependence of
flow measurement capability on yaw angle. The variability of the French probe is confirmed by
visual inspection of the confidence interval plots (in Appendix D) and dispersion analysis plots (in
Appendix G). At the two sites where Matrix B tests were performed on the French probe, no test
team or probe copy effects were detected.
6.9 MODIFIED KIEL PROBE
The modified Kiel probe produced the highest volumetric flow values of all the tested methods at
Lake Hubbard, high load (7.3% above the grand mean) and second highest at DeCordova (3.6% above
the grand mean). However, at Homer City, the modified Kiel probe had the second smallest deviation
from the grand mean (0.3%) of all tested methods. At Lake Hubbard, low load it had the smallest
deviation (2.0%) from the grand mean of all the tested in-stack manual methods. The modified Kiel
probe's CV at Homer City (1.24%) was lowest of any of the tested manual probes. At DeCordova,
the CV (1.39%) was in the middle of the range for manual probes. The probe had the second highest
CV at Lake Hubbard (5.19% at high load and 4.04% at low load). The high variability at Lake
Hubbard is evident in the confidence interval plots (in Appendix D). Some probe copy effect was
detected at Homer City, but not at DeCordova. No test team effect was detected at either site.
6.10 COMPARISON TO ENGINEERING BASELINE
One of the collateral goals of the field study was to reduce the disparity reported by industry between
in-stack and combustion-based calculations of heat rate and flow. To evaluate this reported disparity,
the differences between flow values measured by the in-stack probes and baseline values determined
using the MMBtu method were calculated. At DeCordova, Lake Hubbard high load and low load,
and Homer City, respectively, the following percent differences from the MMBtu method flow values
were found: 0.39%, -1.88%, -3.55%, and -1.44% for the Autoprobes 16-point yaw-nulled; 2.45%,
0.04%, -1.16%, and 1.88% for the Type S probe yaw-nulled; and 0.37%, -2.71%, -5.33%, and -0.86%
for the DAT probe.
6-4
-------
6.11 COMPARISON OF MANUAL AND ELECTRONIC PRESSURE MEASURING DEVICES
From a practical standpoint, differences in pressure readings between manual and electronic pressure
devices were generally small. Where statistically significant differences were detected, the
differences were not consistent for different types of probes and different velocity conditions. When
used with Type S probes, however, manual devices tended to read higher pressures than electronic
transducers in field tests, but read lower than electronic transducers in the wind tunnel tests.
Generally, the differences between the manual and the electronic devices are smaller on average and
less variable in the wind tunnel tests than in the field tests.
6.12 WALL EFFECTS
Wall-effects adjustments to flow generally resulted in a decrease in calculated velocity. The average
percent difference in average velocity across all probe types and field test sites for which complete
wall effects data sets were obtained is -1.72% (s.d. = 0.23%, n = 5 sites). (Complete wall effects data
sets were obtained at DeCordova and Lake Hubbard and at three additional sites that were tested
solely for wall effects.) No significant difference was found among probe types. A small difference
was observed between the percent difference in the average velocities measured in smooth (steel)
stacks (-1.50%) and rough (brick and mortar) stacks (-1.86%). However, the variance of the wall
effects measurements was sufficiently large that these differences were not found to be statistically
significant at the 95% confidence level. As expected, the percent decrease in average velocity
between unadjusted and wall-effects adjusted traverses becomes smaller as the number of points in
the Method 1 traverse increases.
6.13 CALIBRATIONS
6.13.1 One- and Two-Dimensional Probes
All probes were calibrated at North Carolina State University (NCSU) before and after field testing.
Probes were also calibrated at National Institute of Standards and Technology (NIST) after the field
testing was completed. Comparison of the NCSU pre-and post-test results showed that all the post-
test Cp values derived at 60 ft/sec were within ±1.5% of the pre-test values (the values used to
calculate flow in the field tests). The largest differences between pre-test and post-test Cp values
occurred at 30 ft/sec. For all probes, the post-test 30 ft/sec Cp values were higher than the pre-test
coefficients, ranging from 0.4% higher for one Prandtl probe to 5.1% higher for one Type S probe.
Pre- and post-test Cp values obtained at 60 and 90 ft/sec were generally consistent. Even with the
comparatively large changes observed at 30 ft/sec, Cp values averaged over all three velocities
changed by less than 1.5% between pre- and post-test calibrations. Although the Cp values measured
at NIST were generally higher than the corresponding NCSU post-test values, the NIST post-test Cp
values were within ±2.2% of the NCSU values.
6.13.2 Three-Dimensional Probes
DAT Probes
For three of the four DAT probes, differences in velocities calculated between pre- and post-test
NCSU calibrations were less than 2% in the -20 ° to +20 ° pitch angle range. In the pitch angle range
of -10 ° and+10°, which is comparable to the range of pitch angles measured at the three utility stacks
during this field study, the differences in velocity were less than 1 %. For reasons that are unclear, the
velocity differences ranged from 3.35% to 4.18% within the -10° to +10° pitch angle range for the
fourth DAT probe. A comparison of calculated velocities between the NIST and post-test NCSU
calibrations shows that, within the -10 ° to +10 ° pitch angle range, calculated velocities over the three
6-5
-------
nominal wind tunnel velocity settings (i.e., 30,60, and 90 ft/sec) were 1.7% or less for each probe.
In the -30° to +30° pitch angle range, the average percent difference of calculated velocity between
the NIST and post-test NCSU calibrations was less than 3.8% for each probe.
Spherical Probes
For the four original spherical probes, the results of the four sets of calibrations atNCSU showed that
the calibration values varied by up to 5% over all pitch angles. For the second set of probes, the
change in calibration was approximately 4% to 5% within the -10° to +10° pitch angle range. A
comparison of the calculated velocities for the NIST and NCSU post-test spherical probe calibrations
shows that, within the -30° to +30° pitch angle range, the calculated NCSU velocity for one probe was
low relative to NIST (-1.2%). For all other probes, the NCSU calculated velocities were 1.9% to
6.4% higher than those obtained by NIST.
6-6
-------
SECTION 7
RECOMMENDED EQUIPMENT AND PROCEDURES
FOR MEASURING VOLUMETRIC FLOW
A variety of equipment and procedures for measuring volumetric flow was evaluated during this
study. This section describes the equipment and procedural revisions and their intended purpose,
actual field performance, and potential value in improving measurements of volumetric flow.
7.1 BACKGROUND
The recommendations contained in this section include proposed performance standards for the
equipment and procedures used in performing yaw, pitch, and velocity calibrations. The
performance standards are expressed as tolerance limits (e.g., ±2°). Tables 7-1 through 7-3 show
the proposed tolerances for each constituent factor and the resulting cumulative tolerance limits
obtained by summing the tolerances of all constituent factors. The "recommended" performance
specifications represent tolerances that the field tests indicated are achievable through careful
performance of the test procedures using currently available technology.
Two principles were followed in establishing the "recommended" specifications on the constituent
factors:
• Field and laboratory tests had to indicate that the specified error limits were well within the
capabilities of commonly available equipment and typical test personnel.
• The error limits on each constituent factor were set so that the cumulative yaw and pitch
angle tolerances would not exceed 10° and the cumulative velocity calibration tolerance
would not exceed 8.5%. Figure 7-1 shows the test method error resulting from various
errors in angle measurements.
7.2 EQUIPMENT REVISIONS
Recommended equipment revisions based on the field study results include specifications for
physical features of probes and probe components and the use of probe supports and stabilization
devices. Additional recommendations involve the use of digital inclinometers for determining yaw
angles, the use of electronic transducers during probe calibrations in wind tunnels and during field
tests, and cross-sectional area requirements for wind tunnels used for determining probe-specific
calibration coefficients.
7.2.1 Probes and Probe Components
Any probe used to determine the yaw angle of the total velocity vector should have certain physical
specifications, such as effective length of the probe, inside diameter (I.D.) of the pressure tubes, a
probe sheath, and a scribe line.
Physical Dimensions
Maintaining the probe head in a horizontal position is an important consideration in limiting error
in angle and velocity measurements. For probes that measure pitch, the degree of deflection from
horizontal results in an equivalent displacement in pitch with the resulting error in velocity shown
7-1
-------
Table 7-1. Proposed Yaw Angle Tolerances
Factor
Wind tunnel axial flow
Reference scribe line precision
Inclinometer precision
Protractor wheel measurement resolution
Pointer needle resolution
Rotational position check
Yaw angle calibration
Cumulative Tolerance*
Tolerances
±3°
±1°
±1°
±1°
±1°
±1°
±1°
7°-8°
The tolerances associated with the protractor wheel and pointer needle are included
in the cumulative tolerance only when these devices are used for yaw angle
measurement.
Table 7-2. Proposed Pitch Angle Tolerances
Factor
Wind runnel axial flow
Pitch plate resolution
Horizontal straightness of probe
Pitch angle calibration curve
Cumulative Tolerance
Tolerances
±3°
±1°
4°
±2°
10°
Table 7-3. Proposed Velocity Calibration Tolerances
Factor
Wind tunnel velocity drift during calibration
Error due to wind tunnel blockage
Calibration coefficient variation
2D probes
3D probes
Within ±10° pitch range
Beyond ±10° pitch range
Cumulative Tolerance*
Tolerances
2%
0.5%
3%
3%
5%
5.5% - 7.5%
Velocity drift must be factored into the cumulative tolerance when a probe is calibrated
in a wind tunnel where the standard and tested pilots take successive pressure
measurements during calibration, not in tunnels where they can take concurrent
measurements during calibration.
7-2
-------
£-L
ore
§
-I
IB-
B|
CL
2 c
41
i
3
i
e
i
D
P+-
M
»
n
B
I
«
O
5*
•<
cT
o
Resulting Error in Velocity
jn p 01 o
0.
m
> w
HL
to
»*
o
1
P
X
e
1
•'
/ X
/
/ / /
/' .X /
' .-' /
/ / r'
x' •' -'
1 1
\ \ r
• > *
• • i
* £ *
t *, i
i i / ?
V ; t f
M r r * ?
1 i ' / /
\ \ \ i I i .'
i • / * / *
\\iii /
; i ; r ,' *
tf*;.'1',"'
i * / '
•* J * / '
:' * '
• i v
J
/ / / ,' t ; i
' ,.'" ' / ' J '
/ / ' / ' ; '
• # * r ^ ^ *
' ,''.**,''
' .' * ' ' f '
' / ' :' '
» T' r ^ i
/ ,• r ,» t
/ ' * •
* .* *
* : *
* ; t
1 : '
I
!• *
^ "'
; t
' t
Ji *
/
•
* * * t «• > * to
u ro M _> ^ (n o $~n
5 in <5 en o <6g-
-------
in Figure 7-1. In addition, even for probes that do not measure the pitch angle of flow, deflection
from horizontal affects the applicable velocity calibration coefficient and could affect the yaw angle
measurements. For example, wind tunnel tests" that were performed in preparation for the field
studies whose results are reported here showed that, at a pitch displacement of-5°, the calibration
coefficient for a Type S probe changed by as much as 2.4%.
Experience gained during the three field tests suggested a simple equipment requirement for
maintaining the horizontal position of manual probes: the effective length of the probe (coupled with
a cylindrical probe extension, if necessary) should be at least 3 ft longer than the last traverse point
marking on the probe shaft. An extension of this length will enable an operator to maintain the
probe's horizontal stability when it is fully inserted into the stack. Other physical mechanisms are
also available for maintaining the horizontal position of the probe. For example, a comparable
horizontal position could be maintained by a probe inserted into a bushing sleeve installed on a test
port or an automated system that includes an external probe casing with a transport system.
The inside diameter (I.D.) of the pressure port connection tubing should be at least 1/4 in. to reduce
the time required for pressure equilibration. This recommendation stems from the field test at the
DeCordova site, where the response time for one of the DAT probes with 1/16-in. I.D. port
connection tubing was approximately one minute longer than the response times of all other probes,
which had 1/8-in. I.D. diameter connection tubing. To further reduce response time, the tubing that
is used to connect the probe and the pressure-measuring device should be as short as practicable.
An on-site response time check of the measurement system setup may indicate that the connection
tubing should be replaced, if the response time appears to be too slow.
Sheath Requirements for Probes Used to Determine Yaw Angle
Velocity probes, particularly Type S pilot probes, are used by testing firms in a variety of stack test
applications and configurations with other test equipment. Current test method requirements include
design specifications for the geometry of the pitot head and pressure ports, but no specifications are
prescribed for the probe shaft. Although the same type of probe head is used in these applications,
a diverse variety of probe sheath configurations is often used, including cylindrical and rectangular
configurations. In some applications, probes are used without sheaths, and the pitot tubes are simply
welded together to add rigidity to the probe shaft.
Some of these probe shaft configurations are unsuitable for performing yaw-nulling during wind
tunnel calibrations and field tests, because precise rotational movements and measurements are
difficult to make. Two-dimensional (2-D) and 3-D (3-D) probes fabricated for this field test and
others supplied by the test teams comprised a uniform set of probes with cylindrical sheaths that
extended along the full length of the probe shaft. The cylindrical outer sheath provided a surface
suitable for inscribing a yaw-null reference line and for attaching an angle-measuring device. The
sheath also enabled precise rotational movement of the probe during the yaw-nulling procedure.
For yaw-nulling with 2-D and 3-D probes, the physical design of the probe should (1) provide a
surface for inscribing a reference scribe line (i.e., a line corresponding to the probe's yaw-null
rotational position); (2) accommodate attachment of an angle-measuring device to the probe shaft;
and (3) facilitate precise rotational movement of the probe for determining yaw angles. The wind
The Cadmus Group, Inc. 1997, "Flow Reference Method Testing and Analysis: Wind Tunnel Experimental Results,"
EPA/430/R-97-013.
7-4
-------
tunnel and field tests reported here suggest that these capabilities can readily be provided by
requiring probes to include a cylindrical sheath rigidly attached to the probe assembly and enclosing
all pressure lines. The sheath should extend continuously from within 4 in. to 12 in. of the probe
head to the farthest position away from the probe head, where an angle-measuring device may be
attached during use in the field. The sheath should be fabricated of material that can be permanently
scribed with a reference scribe line that is sufficiently precise to indicate rotational changes to within
±1 °. The sheath of the fully assembled probe must be sufficiently rigid and straight at all rotational
positions such that, when one end of the probe shaft is held in a horizontal position, the extent of the
bend in the fully extended probe meets the specifications recommended in section 7.3.2 below.
Other sheath designs (e.g., a rectangular sheath mounted in a gimbal) should be considered if they
can be shown to satisfy the three requirements described in the previous paragraph.
Reference Scribe Line
During the pre-test NCSU calibrations of the 2-D and 3-D probes, a calibration was performed to
demonstrate that the yaw-nulling procedure used by each probe was accurate to within ±2° at the
velocities where a calibration coefficient was determined. Following these checks, a reference
position, representing the true zero yaw position, was marked on each probe shaft. Field test
personnel then used this marking as a point of reference for installing the digital inclinometer.
This experience led to several observations. First, an indication of the zero yaw position should be
permanently marked on the probe, not by visual inspection of the probe, but by yaw-nulling the
probe in a wind tunnel. The wind tunnel tests showed that scribe lines placed on probes by visual
inspection could not reliably meet the ±2° specification, whereas those placed by yaw-nulling could
consistently meet this specification. Second, a permanent line inscribed on the probe shaft is very
useful for lining up the angle-measuring device with the zero yaw position on any probe that will be
used to determine yaw angles of flow. Finally, the zero yaw position that is permanently marked on
the probe should be checked by yaw-nulling the probe in a wind tunnel before its initial use in the
field. Any measured offset from the zero yaw position should be recorded for use in offsetting yaw
measurements in the field. This check should be treated as a calibration record and then verified
during each subsequent calibration.
7.2.2 Probe Supports and Stabilization Devices
When probes are used for determining flow angles, keeping the probe head in a relatively stable
horizontal position is critical. Although little can be done to eliminate vibration of the probe head
at a traverse point in the stack, test personnel can, at a minimum, secure the section of probe
extending outside the test port. Two types of stabilization devices were used in the field tests. At
DeCordova and Lake Hubbard, monorails installed above each port were used to support the probes.
For the Homer City test, which was performed from within an annular space between the stack and
outer shell, stands were provided to the test teams upon which the probe shafts were rested to help
maintain horizontal position during flow angle determinations.
The physical characteristics of each test platform may dictate the most suitable type of stabilization
device. For probes longer than 10 ft, the portion outside the stack should be secured. Three types
of devices are recommended: (1) monorails installed above each port, (2) probe stands on which the
probe shaft may be rested, (3) and bushing sleeves of sufficient length secured to the stack ports to
maintain the probe in a horizontal position. Additionally, testers should place a level or an angle-
7-5
-------
measuring device on the portion of the probe shaft that extends outside of the test port to ensure that
the probe is horizontal before taking readings.
7.2.3 Yaw Angle-measuring Devices
During the field tests, digital inclinometers were used on each of the 2-D and 3-D probes instead of
an analog gauge typically used by testers. The digital device proved superior to the analog gauge for
two reasons. First, the digital device provided greater precision in angle measurements, having a
manufacturer's rated precision of 0.1 °. Second, it was easier to read the digital inclinometer during
the field tests when the probe end and inclinometer extended well beyond the railing of the stack
platform.
Digital inclinometers or a comparable angle-measuring device capable of measuring and displaying
the rotational position of the probe to within ±1 ° should be used with manually operated probes.
The device should be able to be locked into position on the probe shaft and retain its original
physical location on the probe shaft during the full course of field testing. Angle readings should
be recorded to the nearest whole number (i.e., the display readings of 0.1 ° should be ignored).
A collar or block that can be attached to the probe sheath may be required for locking the digital
inclinometer or comparable angle-measuring device into position on the probe shaft. An engineering
diagram of the block collar used in these field tests appears in Appendix H of the DeCordova field
test plan.12
Alternatively, a protractor wheel and pointer assembly can be used to measure yaw angles. The
protractor wheel can be attached to a port opening and set in a fixed rotational position to indicate
the yaw angle position of the probe relative to the longitudinal axis of the stack or duct. A pointer
assembly that includes an indicator needle mounted on a circular collar can be placed on the probe
shaft and be locked into a rotational position on the probe's reference scribe line. The protractor
wheel and pointer assembly must be able to read rotational angles to within 1 °.
7.2.4 Pressure-measuring Devices
The pressure gauge comparison reported in Section 4.4 indicated that, in wind tunnel tests, consistent
differences could not be detected among inclined manometers, Magnehelic® gauges, and electronic
manometers. However, the NCSU principal investigator suggested that differences between the
probe calibrations conducted at NCSU at different times (described in Section 2.2.1 of this report)
could be due in part to a 2% to 3% error in the Magnehelic® gauge readings, particularly at the low
pressures measured at 30 ft/sec. When used to measure velocity pressure in electric utility stacks,
small differences were detectable between the electronic manometers and the other pressure-
measuring devices under turbulent flow conditions, but not under more stable flow conditions (see
Appendix 12 of this report).
Superior data capture, elimination of operator error, and ease and speed of use represent advantages
of electronic manometers coupled with an electronic data capture device for both field testing and
wind tunnel use. However, the absence of consistently detected differences between the electronic
manometers and other pressure-measuring devices in most situations suggests that use of electronic
manometers should be recommended rather than required.
12 The Cadmus Group, Inc., 1997, "Flow Reference Method Testing and Analysis: Field Test Plan, DeCordova Steam Electric
Station," EPA/430/R-97-024.
7-6
-------
7.2.5 Wind Tunnel Cross-sectional Area Requirements
Questions have been raised regarding the quality of calibration factors derived in wind tunnels
having small cross-sectional areas relative to the size of the probe head. During this study,
calibration factors assigned to probes by vendors often differed substantially from the calibration
factors derived in the NCSU wind tunnel tests and subsequently confirmed in the NIST independent
calibrations. A primary cause of this disparity appears to be the size of the wind tunnels used by
probe suppliers to calibrate their equipment.
Wind tunnels used to calibrate velocity probes must provide axial flow within the test section, stable
velocity over time, and consistent velocity measurements made by standard and test probes. They
should be capable of achieving velocities between 20 ft/sec and 100 ft/sec. The cross-sectional area
of the tunnel must be large enough to ensure fully developed flow in the presence of both the tested
probe and standard calibration pitot. To ensure this condition is met, the cross-sectional area of the
test section should be of sufficient size that the volume of the probe head, shaft, and attached devices
satisfies the following condition:
V<0.005(A)15 Eq. 7-1
where:
V = volume of the probe head, shaft, and attached devices for both the test and
calibration probes, if placed simultaneously in the wind tunnel, or the larger of the
test and calibration probes, if placed separately in the wind tunnel; and
A = cross-sectional area of the test section.
Equation 7-1 is a generalized statement of the conditions necessary to ensure that the effect of the
probe blockage on the flow velocity in a wind tunnel will not exceed 0.5%. For the probes used in
the field tests, this inequality will be satisfied when the projected area of the probe head, shaft, and
attached devices does not exceed 4% of the cross-sectional area of the tunnel, and the diameter of
a circular wind tunnel, or width of the shorter side of a rectangular wind tunnel, is at least 12 in.
To further ensure that wind tunnels meet Agency standards, wind tunnel operators should be required
to calibrate Agency-supplied audit probes on a periodic basis and report the results to the Agency.
7.3 PROCEDURAL REVISIONS
Based on the results of this study, several recommendations are being made regarding wind tunnel
operation, probe calibration, and field test set-up and performance.
7.3.1 Wind Tunnel Procedures
Wind tunnel procedures recommended for revision include verifying axial flow, placing scribe lines
on probes, and performing yaw-nulling and velocity calibrations; wind tunnel facility specifications
are also recommended.
Wind Tunnel Axial Flow Verification
The design of the wind tunnel and the location of the test section relative to flow disturbances must
ensure stable, fully developed axial flow patterns at the test point(s) where the probes are calibrated
7-7
-------
and at the point where the calibration pitot tube is positioned for the reference pressure
measurements (if different from the test probe location).
To verify that axial flow exists in these locations within the wind tunnel, a series of yaw- and pitch-
angle measurements must be taken using either a 3-D probe or a 2-D wedge probe. A previously
calibrated 3-D probe would allow the operator to obtain both the yaw- and pitch-angle measurements
using one port in the wind tunnel. Using an uncalibrated 3-D probe or a 2-D wedge probe would
require that measurements be made from two ports, that is, yaw angles would first be measured using
the tested probe port and then measurements would be repeated using the 90° offset port that
provides the pitch angle of flow. These angle measurements should be made at the lowest and
highest velocities at which probes will be calibrated in the tunnel. To ensure that flow in the tunnel
is sufficiently axial to maintain the precision requirements described earlier in this section, all such
angle measurements must be within ±3 ° of 0 °.
Wind Tunnel Velocity Drift Check
To confirm that flow within the wind runnel remains stable, a velocity drift check should be
performed each time a probe is calibrated in the tunnel. The drift check should show that, at each
wind tunnel velocity setting, consecutive velocity pressure measurements taken by the calibration
pitot remain within a predefined limit (e.g., a change of less than or equal to 2%). This check would
not be necessary in cases when concurrent, paired calibration pitot tube and tested probe pressure
measurements are taken.
Placement of Reference Scribe Line
As noted above in section 7.2.1, the zero yaw position should be permanently marked on the probe,
not by visual inspection, but by yaw-nulling the probe in a wind tunnel. Two alternative procedures
could be used to establish the location of this reference scribe line on the probe sheath. In the first
approach, the orientation of the line is established by yaw-nulling the probe in the wind tunnel (i.e.,
yaw angle calibration) and then placing the line at some specified angular position relative to this
determined yaw-null position. The second approach would be to arbitrarily place the reference
scribe line on the sheath, and the offset would then be determined during the yaw angle calibration
in the wind tunnel. This second approach may be preferable because it is likely to be easier to
implement. With either approach, the offset of the line relative to the probe's yaw-null position
would become part of the probe's calibration record and would be used by field test personnel in
determining the correct placement of the angle-measuring device on the probe shaft. Without a
reference scribe line, documented offset, and periodic yaw offset verifications (described below), the
probe should not be used for measuring yaw angles and deriving yaw-adjusted velocity values.
Reference scribe lines may not be necessary for an automated probe system if a reference rotational
position of the probe is built into the probe system design. For these probe systems, a "flat" (or
comparable, clearly identifiable physical characteristic) should be provided on the probe casing or
flange plate to ensure the reference position of the probe assembly remains in a vertical or horizontal
position.
Yaw-nulling Calibration
If probes are used to determine the yaw angle of flow, a quality control check should be implemented
to document the rotational position of the reference scribe line relative to the probe's yaw-null
position to within ±1 °.
7-8
-------
This procedure should be performed before the probe is first used in a field test. It should also be
performed during each subsequent wind tunnel calibration.
Velocity Calibration Procedures
The probe calibration procedures followed during the NCSU and NIST calibrations were based on
the current specifications in Methods 1 and 2 and draft Method 2F. The calibration routine was
performed at three wind tunnel velocity settings of 30, 60, and 90 ft/sec.
A minimum of three replicate measurements was taken at each velocity setting for 2-D probes and
at each velocity and pitch angle setting for 3-D probes. Upon completion of the first set of
calibrations at the three velocities, the wind tunnel was shut down. The operator then repeated the
probe set-up procedures (i.e., verified the alignment of the digital inclinometer to the probe head and
measured the probe tip insertion depth in the tunnel). After the wind tunnel velocity was again set
to nominal, the yaw-nulling procedure and calibration routine were repeated at the three velocities.
The calibration procedures for 2-D and 3-D probes should include the provision to perform the
calibrations at two wind tunnel velocity settings, the expected minimum and maximum velocities
where the probes will be used. The calibration coefficients obtained for 2-D probes should remain
stable to within ±3% over this range or the calibration should have to be repeated at two alternative
velocity settings until the ±3% specification is met. The calibration curves for 3-D probes should
meet similar specifications. At 5° increments over the ±15° pitch range, the percent difference
between the velocity calibration coefficients, F2, obtained at the two chosen nominal settings should
not exceed ±3%. Beyond the ±15° pitch range, the percent difference should not exceed ±5%.
To build adequate quality control into the test method, probe recalibrations should be performed after
a specified time period (e.g., 12 months from its first field use) or number of field tests (e.g., 10).
A recalibration should also be performed whenever routine visual inspection of the probe indicates
a physical change that may affect the previously derived calibration.
Calibration procedures for the 2-D and 3-D probes should be consistent, except for the procedures
specifically affecting the determination of yaw and pitch angles. For example, the calibration
procedure for 3-D probes should be the same as that used for the 2-D probes, except for the
provisions pertaining to pitch-angle orientation during tests and the generation of calibration curves.
The calibration procedures outlined in the unpromulgated 1993 version of draft Method 2F include
the requirement that the 3-D probe be calibrated in the pitch-angle range of-60 to +60°. Because
pitch angles are typically within ±15° at utility stacks, a pitch-angle range of ±15° should suffice
under most conditions. If larger pitch angles are anticipated, then the calibration must include the
expected pitch angle range.
7.3.2 Field Test Procedures
Recommended revisions to field test procedures include procedures to (1) minimize bending of
probe shafts, (2) verify correct placement of the yaw angle-measuring device, (3) account for the yaw
component of flow, (4) account for near-wall velocity decay, (5) account for test port extension into
the stack, and (6) measure velocity at individual traverse points.
7-9
-------
Horizontal Straightness Test
Bends in the probe shaft can affect the measurement of differential pressure and determinations of
yaw and pitch angles. A procedure was implemented in the field tests to ensure that a probe shaft
is at least sufficiently rigid and straight to achieve a basic tolerance for straightness. For probes that
are coupled with a probe extension, this check will verify that the probe head and extension are
properly aligned when assembled. To verify that the probe shaft has not been bent beyond tolerance
in the course of testing, the horizontal straightness test must be performed before and after use in
field testing or wind tunnel calibration.
The procedure is performed on the fully assembled probe (including the probe head and all probe
shaft extensions). The probe is secured in a horizontal position using a stationary support at a point
along the probe shaft approximating the location of the stack entry port when sampling at the farthest
traverse point from the stack wall. The unsupported length of probe should be at least the distance
from the stack entry port to the traverse point farthest from the wall. An angle-measuring device is
used to measure the horizontal position of probe shaft at the secured end. With the probe held in this
position, the angle-measuring device is placed next to the probe head to measure the declination (sag
angle) from horizontal.
Alternatively, the declination can be determined trigonometrically by measuring the distance from
the floor to the centerline of the probe sheath at the controlled end and to the centerline at the probe
tip, that is, at the free end. The sag angle can then be calculated from the relative difference between
the two measurements, which is referred to as the sag distance and the distance along the probe shaft
between the measurement points at the controlled and free ends of the probe.
Based on performance of the horizontal straightness check by the test teams that participated in the
field tests and subsequent checks made on various test probes, the sag angle limits shown in
Table 7-4 were found to be readily achievable.
Rotational Position Calibration
The probe's reference scribe line is permanently marked on the probe shaft, and any offset from the
zero yaw position is documented during the wind tunnel calibration. The reference scribe line and
offset are used by field test personnel to determine the correct placement of the angle-measuring
device on the probe shaft for the field test. A procedure used during the field tests provided
verification that the angle-measuring device, once attached to the probe and aligned with the
reference scribe line (including any offset), correctly indicated the rotational position of the probe
head to within ±2 °.
Table 7-4. Horizontal Straightness Declination Limits
For Horizontal Distance (dp)
Between Controlled and Free End
dp <, 10 ft
10ft. < dp <; 20 ft
dp > 20 ft
Sag Angle (a)
Specification
|a| s2°
|a| s3°
|a| <4°
7-10
-------
With the fully assembled probe held horizontally, direct measurements were made of the rotational
positions of the reference scribe line and the angle-measuring device to verify that the specification
was met. A digital inclinometer was affixed to the probe sheath and locked into position on the
reference scribe line. This inclinometer provided a measure of the rotational position of the
reference scribe line. In the pre-test procedure, this angle was used to align a second digital
inclinometer (i.e., the one attached to the other end of the probe to measure yaw angles during the
field test) in the same rotational orientation as the reference scribe line (to within ±2 °). The post-test
check consisted of repeating the procedure to confirm that the rotational orientation of the digital
inclinometer that was attached to the probe was still within ±2° of the rotational position of the
reference scribe line.
A similar out-of-stack check can be performed on an automated probe system by placing an
inclinometer on a "flat" part of the probe casing or a flange plate to establish a reference vertical or
horizontal position of the probe assembly. The rotational position of the probe head then can be set
to a series of yaw angles, as determined by the controller for the automated system, and checked by
positioning the inclinometer on the side of the probe head.
Experience in the field tests indicated that a ±1 ° standard was readily achievable in the pre-test
check, and a ±2° standard was readily achievable in the post-test check.
Although not experienced during the three field tests conducted for this project, testers may
encounter physical constraints at a test location that prohibit full assembly of the probe and extension
outside of the stack. In circumstances where a probe extension has to be added to the probe while
remaining in the test port, the rotational position check should be performed each time that an
extension is added to the probe and the angle-measuring device is re-positioned on the probe shaft.
Yaw-nulling Procedures
If probes are to be used to account for the yaw angle component of flow in a stack, suitable yaw-
nulling procedures must be defined. For manual probes, the yaw angle is determined by rotating the
probe until a null differential pressure reading is obtained. If impact and static ports are used for
yaw-nulling (e.g., with the Type S probe), the probe head must be rotated 90° back to orient the
impact opening into the direction of flow. In this position, the yaw angle is the rotational angle
between the pitot impact port and the longitudinal axis of the stack, as indicated by the digital
inclinometer. The differential pressure reading is then recorded at this probe rotational position.
3-D probes are yaw-nulled by rotating the probe until zero differential pressure between ports P2 and
P3 is obtained. The yaw angle is then determined using the digital inclinometer.
An automated probe system that uses a curve-fitting routine for determining the yaw-null position
of the probe head should have its yaw-nulling procedure verified in a wind runnel as being equivalent
to the manual procedure described above.
Wall Effects
Methods 1 and 2 do not include procedures that directly account for the decline in velocity near the
stack wall. Currently, the only way to account for the wall effect is to increase the number of
traverse points across the entire stack, rather than to increase the number of traverse points only close
to the wall. A method was used during the field tests for taking a series of measurements close to
7-11
-------
the stack wall that can be factored into the calculation of the average axial stack velocity in order to
account for any velocity decay near the wall.
Using this procedure, the original velocity measurement at the Method 1 traverse point closest to the
stack wall is replaced with a substitute wall effects-adjusted velocity value that captures any
measured velocity decline close to the stack wall. This procedure (1) assumes that the velocity at
the stack wall drops to zero, (2) takes measurements in 1-in. increments starting as close to the wall
as the chosen probe allows and continuing out as far as the inside edge of the Method 1 near-wall
equal-area segment, and (3) algebraically integrates over these measurements. Because wall effects
measurements are made to within 1 in. of the stack wall, it is particularly important that test ports
be adequately sealed to minimize leakage during wall effects tests.
Traverse Point Locations
The test ports at each of the three field test sites extended into the stack by 1 in. or less. This port
lip extension, which could be greater at other sites, must be accounted for when determining the
position of the traverse point markings on the velocity probe.
After accounting for such physical offsets, the probe length necessary to reach each traverse point
should be marked directly on the probe sheath. Before beginning any field test, an out-of-stack
verification should be performed to ensure that these position markings are correct. For manual
probes, these point positions should be marked on the probe shaft and verified by measuring the
distance of each marking from the probe pressure port. The automated probe system can extend the
probe to each of the prescribed point positions, which can be verified by measuring the distance
between the port flange and probe head pressure port.
Calculation of Average Velocity
Unlike draft Method 2F, Method 2 does not derive flow velocity values at individual traverse points.
Instead, to derive the average stack gas velocity, Method 2 averages the differential pressures
obtained at all the traverse points and then uses the resulting average differential pressure to calculate
average stack gas velocity. Such an approach does not allow use of information from yaw-angle
determinations, which require calculation of flow velocity at each individual traverse point.
If yaw angle is to be determined, the approach in Method 2 should be replaced by a procedure that
allows differential pressure to be measured at individual traverse points and flow velocity to be
calculated at each traverse point. The yaw-adjusted velocity is then obtained by multiplying this
calculated impact velocity by the cosine of the yaw angle measured at that traverse point. Finally,
the yaw-adjusted velocity values at all traverse points are averaged to obtain the average stack gas
velocity. Such a procedure was used with the yaw-nulled probes involved in the field study.
7-12
------- |