You are on page 1of 22

Supplemental 

Information: Particle Emission Characteristics of a Gas Turbine with a Double
Annular Combustor 
Adam M. Boies, Marc E. J. Stettler, Jacob J. Swanson, Tyler J. Johnson, Jason S. Olfert,
Mark Johnson, Max L. Eggersdorfer, Theo Rindlisbacher, Jing Wang, Kevin Thomson, Greg
Smallwood, Yura Sevcenco, David Walters, Paul Williams, Joel Corbin, Amewu A. Mensah,
Ramin Dastanpour and Steven N. Rogak

Emission Indices

Black   carbon   emission   indices,   EI(BC),   were   compared   for   three   different   laser   induced
incandescence (LII) instruments on three different lines. The results of measurements on the
FOCA, Annex and SAMPLE III lines are shown in Figure S1, and demonstrate that the
differences in measured EI(BC) among instruments and sample lines is less than the different
EI(BC) at each given thrust setting. For each thrust setting the variability of the CFM56­5B4­
2P EI(BC), indicated by the grey error bars, as measured by each LII overlaps demonstrating
that no distinction can be made by the measurements from individual instruments or lines at a
given thrust. These results establish that reported EI(BC) values using one LII on one line are
not statistically different than using other LII results on other lines.

Figure   S1:   Black   carbon   emission   index,   EI(BC),   for   CFM56­5B4­2P   at   various   thrust
settings as measured by LII on three separate lines shown in Figure 1 for 28 th April. The inset
is a rescaled version of EI(BC) for DAC operation.

The variability for particle number based emission indices (EIn) was low and was omitted
from the main article to allow for better view of symbols. SI Figure 2 below is a reproduction

1
of Figure 4 within the main article with the addition of 5 and 95% variability. The variability
within the system was the greatest source of uncertainty within the system.

Figure S2: Black carbon number emission index, EIn, for CFM56­5B4­2P as measured by
CPCs with D50 cut points of 10 nm (open symbols) and 23 nm (cross and plus sign) on two
separate lines (FOCA and SAMPLE III) for various thrust settings. Error bars represent the
90% variability interval within a given thrust setting.

Air-to-Fuel Ratio

Global air­to­fuel ratio was as determined from undiluted CO2 measurements according to the
method defined by the Office of Aviation Research (2006). The air­to­fuel ratios relative to
the stoichiometric air to fuel ratio (SAFR = 22.72) are shown in Figure S2 for varying thrust
settings. Single pilot combustion occurred during low thrust settings (< ~25% full thrust) and
resulted   in   a   global   lean   combustion.   However,   the   local   stoichiometry   within   the   pilot
combustor is rich (not measured) and only mixes with excess air downstream of the pilot
combustion zone. The rich combustion ensures stability of the flame. During dual annular
combustion the global air­to­fuel ratio drops, and the second combustor operates in a locally
lean­burn condition. Local stoichiometry data was not measured for either pilot combustor or
DAC thrust settings.

2
Figure S3: Global air­to­Fuel (AFR) ratio for varying thrust settings over the range of thrust
settings.

3
Effective Density

The particle effective density was determined as a function of particle mobility in the manner
k Dfm
described by  Johnson et al. (In Press 2014)  where     = 11.92 and     = 2.76. The
effective density of the mean mobility diameter is plotted in Figure S3 as a function of thrust
setting.   The   effective   density   is   seen   to   decrease   from   ~900   kg/m 3  to   ~720   kg/m3  with
increased   thrust   setting   during   pilot   combustion   mode.   When   comparing   these   results   to
Figure 5 where it is shown that particle mobility diameter increases over the same thrust
range, it is apparent that the effective density of particles decreases as the particle diameter
increases. The change in effective density is also known to impact the signal decay from LII
measurements, resulting in longer decay times for particles of greater effective densities. For
DAC   thrust   settings   (>   ~25%   full   thrust)   more   than   90%   of   the   effective   densities   are
determined to be between 800 to 900 kg/m. In almost all cases the effective densities, as
determined   by   using   mobility   measurements   from   the  DMS500   (FOCA,   SAMPLE   and
Gantry)   are   greater   than   the   effective   densities   determined   by   using   the   nanoSMPS
(SAMPLE) and longSMPS (SAMPLE/FOCA).

Figure S4: Number mean aggregate effective density versus thrust as determined by mass­
mobility relations at a range of engine thrusts. The effective density is derived from Eq. 2 and
the mobility measurements shown in Figure 5 from DMS500 (FOCA – asterisk, SAMPLE –
triangle, Gantry ­ Circle), nanoSMPS (SAMPLE – diamond), longSMPS (SAMPLE/FOCA –
square).  

Aggregate Mobility Diameter and Mass

4
The mean aggregate mobility diameter was measured for multiple CPMA­selected aggregate
masses at different thrust settings (circle – 9%, triangle – 17%, square – 21%, star – 24% and
diamond – 31%). The results are shown in Figure S4 for both undenduded (white fill) or
denuded (black fill), where there is a similar power­law relationship observed for all samples.
These results are equivalent to those shown in Johnson et al. (In Press 2014).

Figure S5: Measured mean aggregate mobility for a CPMA-selected aggregate mass. Thrust
settings are represented by symbols (circle – 9%, triangle – 17%, square – 21%, star – 24%
and diamond – 31%), face color indicates whether the samples were undenduded (white) or
denuded (black) and line type indicates whether the samples were collected from the FOCA
(solid) or Annex III line (dashed).

5
Primary Particle Diameter

The volume area equivalent primary particle diameter was determined as described in §2.2,
Eq. 4, which relates the volume area equivalent primary particle diameter within an aggregate
to the measured aggregate mass and mobility. As described in the text accompanying Figure
6,   volume   area   equivalent   primary   particle   diameter   increases   with   aggregate   mobility,
0.8
whereby a power­law relationship of   d va =0.79 d m ±25   encapsulates all but one of the
measured data points. Individual power­law fits are shown in Figure S5, which demonstrate
that higher R2 values (R2 > 0.96 for all but one thrust setting) can be achieved when fitting
D va
individual   thrust   settings.   The   resulting     values   vary   from   0.65   to   0.98   for   the
k va D va
individual fits with   values that vary from 1.39 to 0.44, where larger   correspond
k va
to smaller   values.  

Figure S6: Volume area equivalent primary particle diameter as a function of aggregate
mobility diameter as measured by mass and mobility analysis. Primary particle size is
determined according to Eq. 4. The blue lines correspond to the empirical fit with the power
law form of Eq. 5 for each thrust setting.

6
Mass-Mobility Exponent

7
D fm
The   mass­mobility   exponent,   ,   was   compared   for   a   method   by   which   it   was
determined by effective density data to Eq. 2, and from a method of fitting the data to the
volume   area   equivalent   primary   particle   data.   The   ordinate   represents   the   mass   mobility
Dfm
exponent,  , determined by taking the best fit to effective density data as conducted by
Johnson   et   al.   (In   Press   2014).   The   abscissa   represents   the   mass   mobility   exponent   as
Dva
determined by finding     from a least­squares fit to primary particle data as shown in
D fm=2 Dα −D va (2 Dα −3)
Figure 6 and solving the relation,   (Eq. 5 in main article) with a

=1.069. The difference in the Dfm is small and arises from the differences of conducting
a least squares fit to the same data weighted in different manners. These results confirm that
Dα D fm
using the Eggersdorfer constant,   =1.069, allows for conversion between     and
D va
.

D fm
Figure S7: Mass-mobility exponent, , determined by taking the best fit to effective
density versus mass-mobility exponent as determined by a fit to primary particle data shown
D va
in Figure S5 for and calculated using Eq. 5.

8
Dynamic Shape Factor

Dynamic   shape   factor   may   be   determined   by   comparing   the   measured   mobility   and
aerodynamic   diameters.   AMS­measured   vacuum   aerodynamic   diameter   was   measured   in
parallel   to   mobility   measurements,   see   Figure   1.   The   mean   mobility   diameter   and   mean
vacuum aerodynamic diameter were plotted and compared to lines of constant dynamic shape
χ
factor,  , as defined by DeCarlo et al. (2004). The dynamic shape factors of the measured
aggregates are equal to or less than 1, which is a non­physical result. It was concluded that
insensitivity   of   the   AMS   to   vacuum   aerodynamic   diameters   less   than   50   nm   biased   the
measurements, resulting in mean measured vacuum aerodynamic diameters that were skewed
to larger diameters. Further work is needed to improve the sensitivity of AMS for aggregates
with diameters less than 50 nm in order to make definitive measurements of dynamic shape
factors for gas turbine BC particles.

Figure S8: Vacuum aerodynamic diameter and mobility diameter of aggregates measured at
seven different thrust settings (black diamonds). Lines of constant dynamic shape factor, χ, as
χ
defined by DeCarlo et al. (2004), are shown along with general relations for for
aggregates of less than and greater than 60 primary particles.

TEM Image Processing

A semi-automatic image processing program developed in MATLAB was used for the
analysis of the TEM micrographs (Dastanpour et al., 2014). Grayscale TEM image (Fig. 8-a)
are binarized by setting threshold level for the brightness of the image (Fig. 8-b). Aggregate
projected area, maximum length and width, 2-D gyration radius, and projected area

9
equivalent diameter of the aggregates are measured from the binary image. Primary particles
are sized manually. To enhance the accuracy, large aggregates are cropped into several
sections. Since primary particles are not perfect spheres, the reported diameter is determined
by the mean of the diameter measured in two different directions (Fig. 8-c).

Figure S9: TEM image processing steps: a) Original TEM image; b) Binarized TEM image; 
c) Manual primary particle sizing

Although   manual   measurement   of   the   primary   particle   size   is   time­intensive,   unlike


automatic methods  (Bescond et  al., 2014; Grishin et al.,  2012), it  does  not require prior
information on the size distribution of the monomers in aggregates. Grishin’s method is also
sensitive to model parameters and has to be calibrated for different TEM images frequently.

Line Loss Correction Factor

As shown in Figure S9, the ratio of line loss corrected to measured particle number and
particle mass concentrations were determined for each thrust setting. The particle number line
CF PN
loss correction factors ( ) varied from 4.4 to 1.7 with the highest correction factor at
low thrust setting, and the lowest at the highest thrust setting of pilot combustion mode.
CF PM
Similarly, the particle mass correction factors ( ) were highest (2.2) at the lowest
thrust setting and lowest (1.5) at the highest thrust setting of the pilot combustion mode. The
line loss correction factors for mass (mean of 1.9) were less than the particle number (mean
of 2.9) correction because of the  inherent weighting of mass mobility distributions to larger
CF PN CF PM
particle sizes, thus lessening the impact of small particle losses. The   and 
were determined using the  downstream measured geometric  mean  and a fixed  geometric
σ
mean ( g =1.7). The results indicate that over the range of distributions measured, the
impact of varying geometric mean had little effect (<10% relative error) on the overall line
loss correction factors.

10
Figure S10: UTRC line loss penetration of 25 m line length and line temperature of 160°C. 
The line loss correction factors were determined for the particle number (PN) and particle
mass (PM) for each thrust setting, shown in Figure S11. The line loss correction routine was
applied   to   the   measured   size   distributions,   which   were   in   general   well   characterized   by
σ
single­mode log­normal distributions with geometric standard deviations   g  = 1.73 (90%
CI   [1.65   1.80]).   Correction   factors   are   greatest   for   thrust   settings   with   smaller   particle
diameters. In  all  cases,  particle  number  correction  factors   were  larger  than  particle  mass
correction factors, as particle number distributions are dominated by small particles whereas
particle mass is proportional to the third moment of the particle size distribution. As shown
there   is   little   difference   between   correction   factors   determined   by   using   the   measured
σ
geometric standard deviation as opposed to a constant geometric standard deviation  g  =
1.7.

11
Figure S11: Line loss correction factors for particle number and particle mass at various
thrust settings. Line loss correction factors indicated by solid symbols were determined using
the   respective   measured   geometric   standard   deviation,   while   the   open   symbols   represent
correction   factors   that   were   determined   with   a   constant   geometric   standard   deviation   (
σ g =1.7
).

As shown in Figure S10, the corrected geometric mean mobility diameter is less than the
measured geometric mobility diameter for all distributions due to the greater propensity for
smaller particles to be preferentially lost during transport. The relationship between line loss
corrected   and   measured   geometric   mean   mobility   diameter   is   well   described   by   a   linear
d =1.165 d p−8.8249 nm
relationship,   pc , where the norm of the residuals is 1.36 for the
corrected versus measured geometric mobility diameter. The linear relationship is used to
correct the reported geometric mobility diameter within the results section of the article. 

12
Figure S12: Line loss corrected particle mobility diameter shown versus measured particle
mobility diameter.

13
Engine Conditions

Figure S13: Measured combustor inlet temperature (T3) and engine speed for CFM56­5B4­
2P at various thrust settings.

As shown in Figure S14 the individually measured datapoints for EI(BC), EIn, and geometric 
mean mobility diameter demonstrate the repeatability of the measured metrics at a set thrust 
point (~17%). In general measurements that were taken on the 28th and 30th April, 2015 
agreed well, with deviations within an individual run that were less than the deviation 
between measurements taken on different days. Measurements that were taken on the 29th 
April, 2015 tended to be more highly variable due to deviations in the dilution system, 
affecting the concentration of CO2 measured and calculated dilution factor. During all 
measurements, despite the variability in thrust setpoint, the measured quantity remained 
relatively steady. The geometric mean particle diameter was relatively insensitive to thrust 
setting, but does appear to be affected by the line in which the measurement was taken on. In 
particular the SAMPLE system appeared to have higher small particle loss (<20 nm) which 
had the effect of increasing the geometric mean mobility diameter of particles measured on 
this line.

14
Variability of Particle Measurements

Figure S14: (a) Black carbon mass emission index, EI(BC), (b) number emission index, EI n,
and   (c)   geometric   mean   diameter   as   measured   by   LII,   CPCs   and   DMS/SMPS   systems,
respectively. Data represents individual measurements (~1 Hz) from each system in order to
demonstrate the repeatability of the measurements.
15
Three representative particle size distributions are shown in Figure S15 for representative 
thrust settings. Two settings were during pilot combustion (6% and 17%) and one during 
double annular combustion (31%). The measured size distributions are all well characterized 
by single­mode log­normal distributions, where the individual scans do not deviate greatly 
from the mean measured distribution over the entire setpoint duration. The particle size 
distributions corrected for line loss are a factor of 1.6 ­ 2.1 greater in total concentration than 
the measured distributions, and the geometric mean particle size is 9­18% smaller for 
corrected versus measured geometric mean particle diameter. 

Figure S15: Particle mobility distribution as measured by nano­SMPS on the SAMPLE line
for   (a)   6%,   (b)   17%   and   (c)   31%   thrust   settings   for   30 th  April.   Colored   lines   represent
individual scans, solid black line represents average of individual scans and the dashed line
represents the line­loss corrected particle size distribution. 

16
Measurement Uncertainty

The measurement uncertainties for the various instruments are given in SI Table 1 below. A 
further discussion of the nature of the uncertainty and verification method is given thereafter.

SI Table 1: Measurement uncertainties for various instruments and physical parameters in 
this study. Derived uncertainty determined in accordance with Abernethy et al. (1985).

1
Correspondence with NRC Canada, as determined in APRIDE-4 and APRIDE-5 campaigns.
2(Symonds, 2010)

3(Johnson et al., 2013)

4(Johnson et al., 2013)

5
As shown in SI Table 3 below.
6
TSI and Grimm CPC product manual.
7
ISO 15900:2009

17
8(Eggersdorfer et al., 2012)

9
References given in main text.

Condensation Particle Counter


Measurement of particle size and concentration by differential mobility analysis (DMA) and
condensation particle counting (CPC), respectively, are well-established, NIST-traceable
methods. SI Table 1 shows the counting accuracy for CPCs in single counting and
photometric modes for the model CPCs used in this study. Both manufacturer’s (TSI and
Grimm) estimate accuracy is better than 10% when used in the single-counting mode, as was
always the condition in this work.

SI Table 2: Counting accuracy for CPCs in single-counting and photometric modes. The
concentration range (in cm-3) provided by the manufacturer for single-counting mode is
shown in parenthesis.

Detailed studies of CPC calibration and accuracy have revealed that CPCs can be calibrated
to levels much more precise than manufacturer’s estimates (Fletcher et al., 2009; Giechaskiel
et al., 2010; Owen et al., 2012). For example, Owen et al. (2012) reported the relative
expanded uncertainty of a CPC (coverage factor of k = 2) is 2.8% over the range of about 1
particle·cm−3 to 104 particles·cm−3. However, because CPC counting accuracy during our test
campaign was conducted on a “spot-check” basis and no instrument-specific calibration
curve was used, a more conservative standard uncertainty of 10% is most appropriate to
apply to the concentration data herein.

Differential Mobility Analyzer


Standard sizing accuracy of the DMAs used was not explicitly provided by the manufacturer,
although the 3081 and 3085 DMAs are ISO 15900:2009 compliant and it is stated that
“rigorous peer reviewed uncertainty analyses have been performed indicating TSI’s DMA has
a sizing uncertainty of approximately <2% (TSI, 2015)”. Vasiliou (2005) found that particle
sizing of 3085 and 3081 DMAs was within the NIST-stated uncertainty range of the
measured PSL particles. Mulholland et al. (2006) conducted a detailed uncertainty analysis
for 100 nm and 60 nm standard reference PSL particles. They found relative expanded
uncertainty of a 3081 DMA (coverage factor of k = 2) for a particle diameter of 101.8 nm to
be 1.1 nm and 0.63 nm for a 63.9 nm particles. For smaller particles like the gas turbine soot
aggregates measured here, the uncertainty will be slightly higher due to the effects of
Brownian motion on the DMA transfer function. Kim et al. (2005) estimated the uncertainty
for a 20 nm particle in a 3085 DMA to be ~ 0.7%. As the case with the CPCs, the
manufacturer’s estimate of uncertainty (i.e. ~2%) appears to best represent the uncertainty of
the particle sizes measured for this work.

Laser Induced Incandescence

18
As determined by the LII mass calibration method, the uncertainty of the measurement ties to
the uncertainty of the calibration and then any uncertainties associated with differences 
between the particulate measured during calibration and that measured in engine tests.  This 
latter uncertainty is an ongoing topic in the SAE­E31 community (VARIANT and MANTRA
campaigns, where instrument response differences with source have been observed).  The 
uncertainty of the calibration is principally defined by the uncertainty of the NIOSH method 
which is suggested to be 16.7% from the NIOSH 5040 standard document.  NPL suggests a 
higher value – 25% (95% confidence limit).  From engine test comparisons (A­PRIDE4, A­
PRIDE5), differences less than +/­ 25% have been observed, which is the uncertainty level 
assumed in this study.

Centrifugal Particle Mass Analyzer
The uncertainty in the CPMA measurements were explored in detail by Symonds et al. (2013)
and they determined a 2.8% uncertainty in the CPMA mass-to-charge setpoint at a 95%
confidence interval (CI). The uncertainty in the standard DMS500 measurements were
determined by Symonds (2010) to be 10% in mobility size at a 95% CI.

Differential Mobility Spectrometer


The DMS500 used in the CPMA-DMS system was modified and therefore the standard
DMS500 uncertainties do not apply. The uncertainty of the modified DMS was determined
by comparing it against a DMA and knowing the uncertainty in the DMA measurements
(Mulholland et al., 1999). The results of this comparison was indicated that the mDMS has a
mobility diameter sizing uncertainty of 3.04%-3.35% (at a 95% CI) depending on the particle
mobility size. Further details regarding this comparison and the uncertainty of the mDMS are
shown in the supplemental information of Johnson et al. (2013). Johnson et al. (2013) also
found the uncertainty in the effective particle density measured using the CPMA-mDMS
system was 9.5%-10.4% (at a 95% CI) depending on the particle mobility size.

Transmission Electron Microscopy


Average standard error for TEM primary particle diameter (either dp or dva) is ~2 nm (1.6 
nm). Measured standard deviations and standard errors are reported in SI Table 3. Imaging 
magnification varied from 100k to 300k for the whole range of test points, which typically 
corresponds to ~1 to 0.6 nm/pixel resolution. The image processing procedure takes into 
account the exact magnification. Although these changes in resolution may influence the 
results slightly, almost all images taken from ONE test point (RPM/load) are at the same 
magnification.  

19
SI Table 3: Standard deviations and standard errors for TEM measured primary particle 
diameters.

20
SI Bibliography

Abernethy, R., Benedict, R., Dowdell, R. (1985). ASME measurement uncertainty. Journal of
Fluids Engineering. 107:161­164.

Bescond, A., Yon, J., Ouf, F., Ferry, D., Delhaye, D., Gaffié, D., Coppalle, A., Rozé, C.
(2014). Automated determination  of aggregate primary particle  size distribution  by TEM
image analysis: application to soot. Aerosol Science and Technology. 48:831­841.

Dastanpour, R., Rogak, S. N. (2014). Observations of a Correlation Between Primary Particle
and Aggregate Size for Soot Particles. Aerosol Science and Technology. 48:1043­1049.

DeCarlo, P. F., Slowik, J. G., Worsnop, D. R., Davidovits, P., Jimenez, J. L. (2004). Particle
Morphology   and   Density   Characterization   by   Combined   Mobility   and   Aerodynamic
Diameter Measurements. Part 1: Theory. Aerosol Science and Technology. 38:1185­1205.

Eggersdorfer, M. L., Pratsinis, S. E. (2012). The Structure of Agglomerates Consisting of
Polydisperse Particles. Aerosol Science and Technology. 46:347­353.

Grishin,   I.,   Thomson,   K.,   Migliorini,   F.,   Sloan,   J.   J.   (2012).   Application   of   the   Hough
transform for the automatic determination of soot aggregate morphology. Applied optics.
51:610­620.

Johnson, T. J., Olfert, J. S., Symonds, J. P. R., Johnson, M., Rindlisbacher, T., Swanson, J. J.,
Boies,   A.   M.,   Thomson,   K.,   Smallwood,   G.,   Walters,   D.,   Sevcenco,   Y.,   Crayford,   A.,
Durdina, L., Bahk, Y. K., Brem, B., Wang, J., Dastanpour, R., Rogak, S. N. (In Press 2014).
Effective   density   and   mass­mobility   exponent   of   aircraft   particulate   matter.   Journal   of
Propulsion & Power.

Johnson, T. J., Symonds, J. P., Olfert, J. S. (2013). Mass–mobility measurements using a
centrifugal particle mass analyzer and differential mobility spectrometer. Aerosol Science
and Technology. 47:1215­1225.

Mulholland, G. W., Bryner, N. P., Croarkin, C. (1999). Measurement of the 100 nm NIST
SRM 1963 by differential mobility analysis. Aerosol Science and Technology. 31:39­55.

Office of Aviation Research, (2006). Determination of Fuel/Air Mass Ratios for Jet Fuels at
Their Flash Point Temperatures, in: Transportation, U.S.D.o. (Ed.). DOT/FAA/AR­02/96,
http://www.tc.faa.gov/its/worldpac/techrpt/ar02­96.pdf.

21
Symonds,   J.   (2010).   Calibration   of   Fast   Response   Differential   Mobility   Spectrometers.
National   Physical   Lab.,   Metrology   of   Airborne   Nanoparticles,   Standardisation   and
Applications, London.

22

You might also like