Aircraft-engine particulate matter emissions from conventional 1 and sustainable aviation fuel combustion : comparison of 2 measurement techniques for mass , number , and size 3

aMetrology Research Centre, National Research Council Canada, Ottawa, Ontario, 10 Canada 11 bGerman Aerospace Center (DLR), Institute of Combustion Technology, Stuttgart, 12 Germany 13 cNASA Langley Research Center, Hampton, Virginia, USA 14 dScience Systems and Applications, Inc., Hampton Virginia, USA 15 eDepartment of Chemistry, Missouri University of Science and Technology, Rolla, 16 Missouri, USA 17 fAerodyne Research, Inc., Billerica, Massachusetts, USA 18

0 AMT Feature: short summary (max. 500 characters incl. spaces) 21 The combustion of sustainable aviation fuels in aircraft engines produces 22 particulate matter (PM) emissions with different properties than conventional 23 fuels due to changes in fuel composition. Consequently, the response of various 24 diagnostic instruments to PM emissions may be impacted. We found no significant 25 instrument biases in terms of particle mass, number, and size measurements for 26 conventional and sustainable aviation fuel blends despite large differences in the 27 magnitude of emissions. (Jet A-1) jet fuels were combusted in a V2527-A5 engine, while an additional 37 conventional fuel (JP-8) was combusted in a CFM56-2C1 engine. 38 We evaluated nvPM mass concentration measured by three real-time sampling 39 techniques: photoacoustic spectroscopy, laser-induced incandescence, and the  (Durand et al., 2021;Elser et al., 2019;Lobo et al., 2015aLobo et al., , 2016. Here we present the inter-comparison of real-time measurements of aircraft 165 engine nvPM emissions in terms of physical characteristics such as mass, number, 166 and size distributions using different diagnostic instruments and measurement 167 principles. The nvPM mass emissions were evaluated using three real-time 168 sampling techniques: photoacoustic spectroscopy, the extinction-minus-scattering 169 technique, and laser-induced incandescence (LII), and two alternative 170 measurement techniques widely used in laboratories and on-board aircraft: filter-171 based photometry and PSD integration. We note that one of the photoacoustic 172 instruments and the LII instruments have been demonstrated to be compliant with 173 the ICAO SARP performance specifications. The PM number-based emissions were 174 measured using a condensation particle counter. The PSD characteristics 175 measured by scanning mobility particle sizers and an electrical mobility 176 spectrometer were also compared. The nvPM and total PM emissions were 177 delineated using a thermal denuder and a catalytic stripper. We also report the 178 effect of laser fluence on the laser-induced incandescence of nvPM for SAF 179 combustion as changing carbon nanostructure is known to influence particle light 180 https://doi.org/10.5194/amt-2021-320 Preprint. Discussion started: 3 January 2022 c Author(s) 2022. CC BY 4.0 License. absorption and consequently LII signals, and hence the derived nvPM mass 181 concentration. The impact of fuel composition on PM emissions will be reported 182 separately (Schripp and NDMAX-Team, 2021).  185 In the majority of this work, emissions were sampled from a single IAE 186 V2527-A5 starboard engine of the DLR ATRA aircraft (Airbus A320-232). The 187 engine was operated on two conventional, petroleum jet fuels, referred to as REF3 188 and REF4, and three sustainable aviation fuel blends, referred to as SAF1, SAF2, 189 and SAF3. The abbreviations for the two conventional petroleum fuels are used to 190 avoid confusion with the previous ECLIF campaign (Schripp et al., 2018).

191
A limited number of experiments were also performed with 192 combusted in the starboard CFM56-2C1 engine (#3) of the NASA DC-8 aircraft.

193
Due to limited fuel availability, none of the other five fuels could be combusted in 194 the CFM56-2C1 engine. The properties of the six fuels are summarized in Table 1 208 An extensive suite of aerosol and gas-phase instruments operated by the 209 members of six different institutions were deployed in two different shipping 210 containers to characterize the emissions ( Table 2). The complete emission-211 sampling setup is discussed in companion papers (Anderson and NDMAX-Team, 212 2021; Schripp and NDMAX-Team, 2021). Briefly, emissions were sampled 213 through a probe located 43 m downstream of the starboard engine of the aircraft.

214
The probe was placed in front of a blast fence located on the western side of the   MST instrument suite; the system is compliant with specifications for the 234 standardized nvPM sampling and measurement system (SAE, 2013;SAE, 2018;235 ICAO, 2017) and whose performance has been demonstrated and evaluated in 236 previous studies (Lobo et al., 2015b(Lobo et al., , 2016(Lobo et al., , 2020. Additional instrumentation 237 installed as part of the NARS included a fast electrical mobility spectrometer of the instruments installed inside these two containers are listed in Table 3. A suite of gaseous emissions was measured in this study, as summarized in Table   243 2. The CO2 measurements from the NASA LI-COR 7000 were in good agreement   TDs are commonly used on-board aircraft for measuring nvPM number 267 concentration and size distributions (Clarke, 1991;Moore et al., 2017) and have 268 been shown to effectively evaporate nucleation and accumulation mode sulfate   8.0 ± 0.7 m 2 g -1 at 550 nm. In this study, we have used the Bond and Bergstrom 286 value of 7.5 m 2 g -1 for consistency with earlier work and instrument software.

287
These values are assumed to vary inversely with wavelength, with an Angstrom 288 (power) exponent of 1; for example, the 660 nm CAPS PMSSA monitor data were 289 processed with a MAC of 7.5 m 2 g -1 × (550 nm / 660 nm) 1 = 6.5 m 2 g -1 .

315
During on-site calibration of the PAX using graphitic Aquadag nanoparticles, the 316 PAX signals were observed to drifted slowly upwards after each baseline. We were 317 nevertheless able to obtain useful data by configuring the PAX to auto-baseline 318 every 180 seconds, and only using the first 15 seconds of measurements after each 319 baseline. After the campaign, it was found that a component of the circuit board 320 was damaged during the initial shipment. In spite of this electrical problem, the 321 PAX data do not represent outliers in the following analysis. correct for filter loading effects (which do not require independent measurements) 334 and may also be corrected for light attenuation due to scattering rather than 335 absorption (which requires an independent nephelometer measurement) 336 (Virkkula, 2010). Other sources of error include nonlinearities due to size-337 dependent penetration of particles into the filter media and the evaporation of 338 volatile species over time (Lack et al., 2014;Nakayama et al., 2010). We note that 339 the TAP automatically advances its filter when its transmission drops below 80%, 340 whereas the PSAP requires a manual filter change. The PSAP filter was therefore 341 changed manually before each set of experiments herein, to ensure that its filter 342 transmission remained above 80% during all measurements.

343
Finally, three Artium LII 300 (Artium Technologies) instruments measured rBC,   Due to the prevailing crosswind mentioned above, unstable CO2 concentrations 375 occurred during from some test points at the idle engine thrust condition. These 376 unstable conditions were identified and filtered using two separate methods. In 377 the first method, the SMPS PSDs were inspected for reproducibility. In the second 378 method, an algorithm was used to reject any test points with CO2 uncertainties 379 greater than 50%, CO2 signals less than a factor of ten greater than uncertainty, or 380 CO2 signals less than 20% above baseline. We found that the first method rejected 381 all of the points rejected by the algorithm, in addition to a few additional points.

382
The analysis presented uses the first method.

383
Third, all data were arithmetically averaged over the test point periods defined in 384   Table S1. For each instrument, the averaging periods were refined by inspection of    Table S1.  shows that the relationship for EInum is less clear, with a slight increase at 459 moderate N1 followed by a greater decrease at high N1. As discussed below 460 (Section 4.2.3), the higher EIm at higher N1 thrust was associated with larger 461 particle sizes, and therefore smaller penetration-function corrections (Section  of each instrument in the sampling system, as specified in Table 2. The magnitude 478 of each correction is given in Table S1.  and showed that the size range measured by these two instruments was 496 insufficient to capture the full PSD for the CFM56-2C1 engine data at 22% N1 as 497 well as 63% N1. The < 10nm mode was not as prominent in the V2527-A5 498 engine exhaust at any thrust, although some evidence was observed for it (e.g.

500
Since the CFM56-2C1-with-JP-8 data were strongly influenced by a nucleation 501 mode, and were therefore not well described by the GMD and GSD of the data, 502 these measurements have been omitted from all subsequent PSD analysis in this 503 manuscript. Bimodal fits to the data were not possible as the nucleation mode was 504 not captured by our size distributions. However, the nvPM mass measurements 505 are much less sensitive to these small particles (Hinds, 1999) and have therefore 506 been retained. PSDs from all instruments, test points, and fuels from both the 507 CFM56-2C1 and V2527-A5 engines are included in the supplement. was not operated behind a volatile particle remover (CS or TD). Moreover, the 517 inversion of DMS500 data requires more assumptions about the particle size 518 distribution than the analogous SMPS calculation. Either volatiles or this inversion 519 procedure may have caused the 10% larger GSDs observed for the DMS500 for 520 some data (some measurements with GMDs over 35 nm) relative to the SMPS.

521
Since volatiles would affect both GMD and GSD, but we primarily observed 522 discrepancies in the DMS500 GSD, we suggest that the inversion was the major 523 source of bias in these data.  In spite of these trends in GMD and GSD, the PSD measurements agreed to within 537 20% ( Figure 7a) for nvPM GMDs and within 5% for nvPM GSDs (Figure 7b).

538
Furthermore, these measurements are consistent with previous measurements by 539 Lobo et al. (2015c), as illustrated by the line in Figure 6, which reproduces the 540 polynomial best-fit line reported by those authors.  shows that all instruments agreed to within a factor of 2. The APC and DMS500 546 nvPM EInum were both typically higher than the two similar SMPSs. The APC has a 547 50% efficiency at its cut-off diameter of 10 nm, reaching 100% efficiency above 548 this size and 0% below it. Therefore, relative to the SMPSs, which measured down 549 to approximately 10 nm with 100% efficiency, the APC should measure lower than 550 the SMPSs since it will underperform at sizes close to 10 nm. (This expectation 551 requires that there are no particles present above the SMPS upper detection limit 552 of 280 nm in our study, which was verified by our PSD analysis in Section 3 and 553 Table 2). However, the APC measured approximately 50% larger nvPM EInum 554 under all conditions, and our measured PSDs rule out the possibility that 50% of 555 particles were not seen by the SMPS. Therefore, we attribute the difference 556 between APC and SMPS results to uncertainties in the APC or SMPS penetration 557 correction functions. Since the two SMPSs agreed, the APC measurements were 558 likely overcorrected when the SARP correction procedures were applied. 559 We also attribute the larger nvPM EInum measured by the DMS500 to the same 560 cause; to which a similar penetration function as the APC applies (Section 3.4.2).  aerosol absorption or to aviation nvPM, we consider their accuracy as greater than 602 the instruments in Figure 8b and consider departures from the 1:1 line as due to 603 inaccuracy.

604
Most of the instruments in Figure 8b were accurate to within 30% of the reference, 605 similar to Figure 8a, with the exception of the CS-SMPS and PSAP. This is 606 summarized in Table 3, which shows the mean ratios of all data except engine idle 607 (23% N1) with the geometric mean. Table 3  PSDs, which, however, were not corrected for diffusional particle loss (Lobo et al., 624 2015b(Lobo et al., 624 , 2020. Careful measurement of the penetration functions used in these 625 calculations would be required to confirm our interpretation.  The PSAP, on the other hand, showed much greater variability, with an RSD of 36% 634 (Table 3). This is substantially higher than the variability reported by a laboratory the measurements become somewhat more scattered at smaller sizes for the SAF1 657 data set, where signal to noise is lower (GMD and EIm were correlated, see the 658 below discussion of Figure 12). Figure 10b includes the size-dependent PSAP correcting PSAP and TAP measurements by the ratios shown in Table 3, which 674 represent the ratio between the calibrated aerosol-phase nvPM mass 675 measurements and the previously uncalibrated PSAP and TAP measurements, for 676 data above 25 mg kgfuel -1 and N1 > 40%.

Instrument performance for fuels with different composition
678 Figure 11 shows a category plot of the ratio EIm/mean-EIm (that is, the ordinate of 679 Figure 9) for the different instruments. Data below 100 mg / kgfuel have been 680 excluded as this ratio reflects only noise in that region (Figure 9). The symbols  Figure 11 shows that no substantial difference can be seen for these instruments 686 for the nvPM EIm for fuels with different composition; the spread in the data for a 687 given fuel is larger than the difference between fuels. Outliers tend to be associated 688 with low N1 (small symbols). Because low N1 corresponds to both lower 689 concentrations (lower signal-to-noise) and lower exhaust velocities relative to 690 ambient wind speeds, these outliers are not surprising.

691
The instruments in Figure 11 show a linear response to nvPM mass and operate on uniquely sensitive to changes in particle size over the observed range, since EIm 694 was correlated with GMD ( Figure 12), as is typical of aviation engines (Saffaripour 695 et al., 2020). We note that the response of all of these instruments is proportional 696 to the MAC of the sample, so that it remains possible that the sample MAC changed 697 with GMD or EIm. 713 Figure 13a illustrates the experiment we performed to test this hypothesis. The 714 figure presents data for SAF1 only; results for other fuels were similar. One 715 "reduced-fluence" LII 300 was programmed to change its Q-switch delay from 716 140 μs to 240 μs, with a randomized order. In this experiment, lower Q-switch 717 delays corresponded to higher laser fluence; the lowest Q-switch delay was the 718 optimal one for this system. Another "reference" LII 300 operated with no change 719 to its Q-switch delay. Figure 13a shows that the reduced-fluence LII reported lower 720 mass concentrations when its Q-switch delay was increased, but returned to the 721 expected values when its Q-switch delay was reduced.

722
We defined as the ratio of nvPM mass concentrations reported by the reduced-

790
RML and AF are employed by ARI, which produces the CAPS PMSSA commercially.

791
ZY was employed by ARI at the time of the study.      instruments is poorer at EIm < 100 mg/kgfuel, which corresponds to an approximate concentration of 10 μg m -3 (the exact conversion factor varies 1127 with CO2 concentration and fuel properties) and close to the limit of detection for most instruments.