9
ILASS Americas 28th Annual Conference on Liquid Atomization and Spray Systems, Dearborn, MI, May 2016 High Resolution X-Ray Tomography of Injection Nozzles K.E. Matusik * , D.J. Duke, A.B. Swantek, and C.F. Powell Center for Transportation Research Argonne National Laboratory Lemont, Illinois 60439 USA A.L. Kastengren X-Ray Sciences Division, Advanced Photon Source Argonne National Laboratory Abstract Tomographic imaging of gasoline and diesel injection nozzles has important applications in industry and research. At high spatial resolution, tomography provides dimensional metrology capable of exposing ma- chining tolerances as well as the effect of nozzle geometry on observed fuel spray behavior. Tracking the three dimensional (3D) structure of the nozzle throughout its lifetime can also reveal any changes to the fuel’s flow path caused by deposit formation and cavitation erosion. Additionally, isosurface visualization of tomo- graphic reconstructions provides true 3D nozzle geometry for mesh generation in computational simulations. In order to make tomography a viable tool for such applications, the technique must produce images at high spatial resolution with relatively low computational demand. Recent upgrades to the 7–BM beamline at the Advanced Photon Source at Argonne National Laboratory now allow us to perform x–ray tomography of fuel injectors with 3 micron spatial resolution. The large energy and flux of synchrotron radiation permits high contrast imaging of the nozzle interior with minimal reconstruction artifacts. Additionally, TomoPy, a parallelizable high performance Python toolbox developed for use at synchrotron facilities, maintains prac- tical computational time for data processing. We present a brief overview of the tomography process and its capabilities, complemented by a sample of reconstruction results of an eight–hole direct gasoline injection nozzle. The 3D nozzle reconstruction exposes micron–scale features with high fidelity and requires minimal manual post–processing to create a 3D isosurface rendering fit for use in CFD meshing. * Corresponding Author: [email protected]

ILASS Americas 28th Annual Conference on Liquid ... · 1.17 m with a 5x objective lens and a eld of view (FOV) of 2.25 1.4 mm. The smallest resolvable feature, found by imaging an

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

Page 1: ILASS Americas 28th Annual Conference on Liquid ... · 1.17 m with a 5x objective lens and a eld of view (FOV) of 2.25 1.4 mm. The smallest resolvable feature, found by imaging an

ILASS Americas 28th Annual Conference on Liquid Atomization and Spray Systems, Dearborn, MI, May 2016

High Resolution X-Ray Tomography of Injection Nozzles

K.E. Matusik∗, D.J. Duke, A.B. Swantek, and C.F. PowellCenter for Transportation Research

Argonne National LaboratoryLemont, Illinois 60439 USA

A.L. KastengrenX-Ray Sciences Division, Advanced Photon Source

Argonne National Laboratory

AbstractTomographic imaging of gasoline and diesel injection nozzles has important applications in industry andresearch. At high spatial resolution, tomography provides dimensional metrology capable of exposing ma-chining tolerances as well as the effect of nozzle geometry on observed fuel spray behavior. Tracking the threedimensional (3D) structure of the nozzle throughout its lifetime can also reveal any changes to the fuel’sflow path caused by deposit formation and cavitation erosion. Additionally, isosurface visualization of tomo-graphic reconstructions provides true 3D nozzle geometry for mesh generation in computational simulations.In order to make tomography a viable tool for such applications, the technique must produce images at highspatial resolution with relatively low computational demand. Recent upgrades to the 7–BM beamline at theAdvanced Photon Source at Argonne National Laboratory now allow us to perform x–ray tomography offuel injectors with 3 micron spatial resolution. The large energy and flux of synchrotron radiation permitshigh contrast imaging of the nozzle interior with minimal reconstruction artifacts. Additionally, TomoPy, aparallelizable high performance Python toolbox developed for use at synchrotron facilities, maintains prac-tical computational time for data processing. We present a brief overview of the tomography process and itscapabilities, complemented by a sample of reconstruction results of an eight–hole direct gasoline injectionnozzle. The 3D nozzle reconstruction exposes micron–scale features with high fidelity and requires minimalmanual post–processing to create a 3D isosurface rendering fit for use in CFD meshing.

∗Corresponding Author: [email protected]

Page 2: ILASS Americas 28th Annual Conference on Liquid ... · 1.17 m with a 5x objective lens and a eld of view (FOV) of 2.25 1.4 mm. The smallest resolvable feature, found by imaging an

Introduction

A thorough understanding of the fuel spray be-havior in combustion engines can guide developmentof higher efficiency ignition processes that also mit-igate emissions. The fuel spray profile is thoughtto be affected by the internal flow in the injectionnozzle, and thus a function of its dimensions. Tothis end, numerous studies have been performed toelucidate the link between injection nozzle geome-try and its effect on the corresponding internal andnear nozzle fuel spray distribution [1, 2, 3]. Previ-ous work by Kastengren et al. [4] provides a rigorousinvestigation of the geometries of a set of four single–hole diesel injection nozzles provided by the EngineCombustion Network using multiple diagnostic tech-niques. The authors found that the measured geom-etry not only deviated appreciably from its targetdimensions, but also varied within the injector set.

X–ray tomography measurements by Duke et al.[5] of three six–hole gasoline direct injection (GDI)nozzles revealed hole–to–hole variations in the spraystructure. Complementary spray and nozzle tomog-raphy measurements in Strek et al. [6] of an eight–hole GDI injector found 7% variability in hole areas,as well as significant plume–to–plume mass flux vari-ations. Numerical simulations verified that inlet–hole diameter substantially affected the mass flowrate in each plume. Additionally, the inlet cornerradius between holes was found to vary by a factorof two, which is expected to affect in–hole cavita-tion. Further experimental work investigating thespray structure as well as the flame and soot in-tensity of two GDI nozzles concluded that manu-facturing technique (electrical discharge machiningvs. laser drilled) and internal geometry (specificallythe L/D ratio) have marked effects on fuel spraycollapse and in–cylinder soot formation [7]. Thus,knowledge of the critical nozzle dimensions such ashole inlet and outlet diameters, inlet corner radius,hole length, and drill angle are precursors for fullyunderstanding the corresponding fuel flowfield, andby extension its effect on the combustion process.

Apart from providing insight into the relation-ship between nozzle geometry and fuel spray behav-ior, a high fidelity three dimensional (3D) modelof an injection nozzle serves further purpose. Fullcharacterization of the nozzle provides flow model-ers with more accurate boundary conditions, allow-ing for increasingly reliable comparison of experi-mental and numerical results. Additionally, the 3Dmetrology reveals machining tolerances, and as suchprovides error bounds on the repeatability and com-parability of datasets obtained from measurementsof nominally duplicate injectors. Lastly, imaging an

injection nozzle during various stages throughout itslifetime of use exposes potential changes to the fuel’sflow path that may be caused by cavitation erosionor deposit formation. A timeline of these events canprovide a gauge of the expected level of correlationbetween datasets that were obtained at varying de-grees of nozzle wear.

To date, obtaining a high fidelity 3D representa-tion of the injection nozzle in a non–invasive manneris somewhat esoteric. Molding has been previouslyused to quantify the nozzle hole dimensions, but thetechnique is limited to the hole exit [8]. Computedtomography (CT), with its origins in medical diag-nostics, makes use of penetrating waves to image thefull 3D structure of a sample by recording its two di-mensional (2D) radiographic projections for a widerange of angles [9]. This technique is used by sev-eral commercial imaging companies to provide CTscans as a service. However, because the prevail-ing wave source used in industry is an x–ray tube,the inherent angular divergence of the beam gener-ally limits the spatial resolution to approximately25 µm. Third generation synchrotron sources, in-cluding the Advanced Photon Source (APS) at Ar-gonne National Laboratory and the European Syn-chrotron Radiation Facility (ESRF), both maintaindedicated microtomography facilities that specializein imaging relatively low density materials [10, 11].While the spatial resolution surpasses that offeredby a benchtop x–ray source, previous experience atthese facilities with imaging a high density mate-rial such as a steel nozzle generally produces imageswith poor contrast, specifically in regions above theinjector nozzle tip where the injector body thicknesssubstantially increases.

Motivated by the need for high resolution 3Drenderings of both diesel and GDI injection noz-zles, the current work details an efficient pipeline forperforming tomography measurements and obtain-ing an isosurface, or 3D representation of the nozzlewall. The isosurface facilitates visualization of thenozzle structure, and creates an end product for in-tegration into computational fluid dynamics (CFD)simulations.

Experimental Method

Tomography measurements were performed atthe 7–BM beamline at the APS at Argonne NationalLaboratory [12]. A recent upgrade to the beam-line has enabled access to the unfiltered broadband(white) beam from the bending magnet. Figure 1plots the power spectrum of the white beam as afunction of photon energy before and after filteringmethods are applied. The higher range of photon en-

2

Page 3: ILASS Americas 28th Annual Conference on Liquid ... · 1.17 m with a 5x objective lens and a eld of view (FOV) of 2.25 1.4 mm. The smallest resolvable feature, found by imaging an

ergies provided by the white beam offers more pene-trating x–rays that partially transmit through densematerial such as the body of a steel injection nozzle,which generally begins to partially transmit x–rayswith photon energies above 40 keV. A schematic ofthe tomography imaging setup is given in Figure 2.

Figure 1: Power spectrum of the unfiltered broad-band (white) beam used for tomography measure-ments. The filtered beam passes through two metalfilters to absorb low energy components upstream ofthe injection nozzle.

To image the injection nozzle, the unfilteredbeam first passes through a 250 µm copper filterand a subsequent 25 µm molybdenum filter in orderto absorb low energy components. The absorptionof these x–rays reduces heat load on the nozzle bodyand minimizes artifacts associated with the poly-chromaticity of the beam. To further reduce heatload, a rotating optical chopper located downstreamof the filters is synced with the framegrabber so thatthe nozzle is exposed to the beam solely during im-age capture with 10% duty cycle. The injection noz-zle is mounted on a rotation stage with its tip in thebeam path. The x–rays that pass through the nozzleare absorbed by a LuAG scintillator and re–emittedas visible light. The resulting projection is magni-fied using an objective lens and recorded with anoff–the–shelf camera to a desktop computer. Leadshielding is placed behind the optics to reduce scat-tered radiation for the purpose of improving over-all image quality. The achievable pixel resolution is1.17 µm with a 5x objective lens and a field of view(FOV) of 2.25 × 1.4 mm. The smallest resolvablefeature, found by imaging an Xradia resolution tar-get in the tomography setup, is 1.5 µm, governedby the diffraction limit of the microscope lens. TheFOV may be increased at the expense of spatial res-

olution, e.g. for a 2x objective, the FOV is 5.6 × 3.5with a corresponding pixel resolution of 2.93 µm.

The imaging workflow begins with recording aset of projections of the nozzle as it is rotated from0 to 180◦. At each sampled angle, a series of imagesis averaged in order to improve the signal–to–noiseratio (SNR) and extend the dynamic range of the im-ages. Two additional sets of images are recorded forthe purpose of intensity normalization: a white–fieldmeasurement with the nozzle outside of the FOV,as well as a dark–field measurement with the x–rayshutter closed. The normalized image set is pre–processed to remove rogue high energy photons inthe image.

The reconstruction algorithm used for all to-mography processing is TomoPy, an in–house open–source Python code developed at the APS [13]. Thismodular software performs various user–selectedprocessing steps that are tailored to the quality ofthe image set, including a smoothing filter to min-imize ring artifacts caused by imperfections on thedetector, a phase retrieval algorithm, and a rota-tional center locator. TomoPy also offers a varietyof reconstruction algorithms; the most efficient forthe purpose of high resolution tomography is a di-rect Fourier–based method called Gridrec [14]. Thismethod makes use of the Fourier slice theorem tobuild a 2D Fourier space from the transforms ofparallel projections of the nozzle [15]. The recon-structed image is found by calculating the inversetransform of the 2D Fourier space after samplingback onto a Cartesian grid [16]. Prior to moving intothe frequency domain, the data is either constant–value or elliptic padded to prevent both aliasingand distortions due to the outer part of the noz-zle being cropped out of the FOV. On account ofits implementation of fast Fourier transforms, thereconstruction algorithm is computationally less ex-pensive than conventional filtered back–projectiontechniques, while also maintaining high accuracyfor adequately sampled data (i.e. sampling thatmeets the Nyquist criterion). Reconstructed imagespass through a ring removal algorithm and total–variation denoising algorithm to prime the stack forisosurfacing [17]. The isosurface is created in Seg3D,an open–source software for volumetric segmenta-tion and visualization [18].

Image Quality and Artifacts

In order to obtain high contrast images withoutbeing heavily reliant on computational techniquesthat tend to globally coarsen the spatial resolution,data acquisition is optimized from the onset of theimaging pipeline. Apart from increasing the SNR

3

Page 4: ILASS Americas 28th Annual Conference on Liquid ... · 1.17 m with a 5x objective lens and a eld of view (FOV) of 2.25 1.4 mm. The smallest resolvable feature, found by imaging an

Figure 2: Schematic of the experimental setup used for nozzle tomography measurements.

(a) (b)

Figure 4: A reconstructed slice of an eight–hole GDI nozzle (a) without beam hardening correction and (b)with beam hardening correction.

and dynamic range, recording a set of images ateach sampled angle provides sufficient data pointsto remove the footprint of scattered radiation onthe recorded image set without loss to spatial res-olution. Anomalous high energy photons that hitthe camera chip directly create relatively bright pix-els, or zingers, on the image. The zinger locationvaries from image to image, and must be removedfrom the full data set to prevent streaking artifactsin the reconstruction. The most effective way of re-moving zingers is to physically minimize their fre-quency by introducing sufficient shielding near theoptical setup that will absorb the scattered radia-tion before it interacts with the CMOS chip. Thecurrent imaging setup achieves this with a 1 cmblock of lead placed along the beam path down-stream of the optics. Figure 3 illustrates the effec-tiveness of proper shielding methods by comparing asingle image acquired at the 7–BM beamline to thatrecorded at the tomography–dedicated APS beam-line [10], where sufficient shielding had not been im-plemented. Any residual zingers that are not ab-sorbed by the lead shielding are cleaned up compu-tationally by making use of the recorded image setat each angular position. The zingers are detected in

(a) (b)

Figure 3: Section of a raw tomography imagerecorded at (a) APS’s general purpose tomographybeamline and (b) at APS 7–BM, exposing the fre-quency of zingers in the subdomain.

each image by applying a threshold filter, and sub-sequently removed by replacing each intensity valuewith the median of the pixel intensities in the re-maining images that do not contain a zinger at thesame location. In this sense, the zinger removal al-gorithm is a temporal rather than spatial filter, thuspreserving information in the image. In previous to-mography pursuits, only one image was recorded at

4

Page 5: ILASS Americas 28th Annual Conference on Liquid ... · 1.17 m with a 5x objective lens and a eld of view (FOV) of 2.25 1.4 mm. The smallest resolvable feature, found by imaging an

each sampled angle, requiring the use of a median fil-ter. As evidenced in Figure 3a, spatially filtering ateach zinger location replaces a significant fraction ofthe raw information with artificial data, ultimatelycompromising the overall resolution.

An inevitable property of the broadband whitebeam is that its lower energy, i.e. softer, com-ponents are preferentially attenuated as the beampasses through the sample, shifting the spectrum tohigher energies, and consequently “hardening” thebeam. Figure 5 plots the transmission as a functionof the path length of iron, which is used as a proxyfor steel, on a semi-logarithmic scale.

Figure 5: Transmission as a function of path lengthof iron on a semi-logarithmic scale. Ideally, trans-mission is proportional to the path length of thebeam through the injection nozzle. Due to beamhardening effects, this linear relationship no longerholds, specifically for long path length.

In this scenario, the transmission no longerobeys the characteristic linear decay with increas-ing path length of the monochromatic beam. In-stead, because the leftover higher energy x–rays areharder to absorb, there is an increase in the fluxof photons that reach the detector downstream ofthe nozzle, i.e. the transmission becomes overpre-dicted. This increased intensity manifests itself asdarkened regions in the reconstructed image, mostevident on the thickest parts of the nozzle body. Afirst–order mitigation to lessen this artifact is ap-plied by physically filtering the softer radiation up-stream of the sample with metal filters that pref-erentially absorb lower energy x–rays. A more ro-bust technique is done computationally. If the noz-

zle material is approximated as being composed ofa homogeneous material (i.e. iron), its path lengthas a function of attenuation is known for the whitebeam spectrum [19]. Therefore, the raw transmis-sion data may be transformed to path length of ironusing a lookup table to correct the effect of beamhardening. Figure 4 shows a single reconstructedimage of the nozzle before and after a beam harden-ing correction is applied. The distinct reduction inintensity in the upper part of Figure 4a correspondsto a thicker section of the injector body, where beamhardening artifacts are amplified. Computationallyadjusting for this nonlinearity between attenuationand measured projection values results in more uni-form intensities, as would be expected in a materialof a single density. The intensity correction substan-tially simplifies thresholding to separate the nozzlebody from the surrounding air, which is critical foreffective isosurface implementation.

X–rays exhibit wavelike behavior, and as such,undergo refraction as they pass between mediums,resulting in a phase shift of the waves. This shift isvisualized as a variation in intensity, creating darkand bright fringes on either side of the material inter-face that ultimately broaden the edges of a feature.Phase effects are minimized by physically constrain-ing the separation between the scintillator and noz-zle to reduce the propagation distance of the phase–shifted x–rays. Phase retrieval algorithms may beimplemented to correct for these effects, but univer-sally assume beam coherence and monochromaticity,properties which are not approximated well by thewhite beam spectrum. In general, phase retrieval al-gorithms come at a cost to the spatial resolution ofthe dataset and it is therefore best practice to avoidtheir use when possible.

Results and Discussion

As an example of the capability of tomographyat the APS 7–BM beamline, we present a tomo-graphic reconstruction of a Delphi eight–hole GDInozzle. The injector is one of a set of nominallyidentical GDI injectors used for measurements andmodeling as part of the Engine Combustion Network(ECN) Spray G initiative, and was thus manufac-tured with the specific intent of being standardizedand repeatable. Projections of the nozzle were ac-quired by rotating the injector from 0 to 180◦ andrecording 20 images at each angle for 1,801 angles at6 ms exposure. After averaging and applying zingerand ring removal algorithms, the images were con-verted to 32–bit depth for reconstruction. On aver-age, the reconstruction algorithm runs for approxi-mately 30 min on workstation–level computer hard-

5

Page 6: ILASS Americas 28th Annual Conference on Liquid ... · 1.17 m with a 5x objective lens and a eld of view (FOV) of 2.25 1.4 mm. The smallest resolvable feature, found by imaging an

(a) (b)

(c) (d)

Figure 6: Tomographic reconstructions from various institutions of one of eight holes of a Delphi GDIinjection nozzle. The image in (a) is a reconstruction from Northstar Imaging, a commercial benchtop x–rayimaging company, while (b)-(c) show reconstructions from data obtained at the ESRF, the general purposetomography beamline at the APS, and APS 7–BM, respectively.

(a) (b)

Figure 7: (a) Cut–away through the isosurface of a Delphi eight–hole GDI nozzle, revealing the surface finishon the nozzle body and check ball, as well as inclusions on the inlet corner of the hole. (b) Two inclusionson the inlet corner of the nozzle holes.

ware with 16 cores and 48 GB of RAM. The bottle-neck generally lies in the denoising step, which takesO(hrs) to finish; in future endeavors, this algorithmwill be parallelized. Figure 6 shows tomography re-sults of a single hole of the GDI nozzle with com-parison to results from three additional institutionswho likewise imaged the Delphi injector.

The hole reconstructions display varying degrees

of spatial resolution and image contrast. Whilethe benchtop x–ray source is limited in resolution,the reconstructed image maintains high contrast be-tween the air and nozzle. Alternatively, the ESRFfacility is capable of sub–micron resolution in re-gions of low steel thickness such as along the coun-terbore, but achieves relatively lower SNR in thehole and check ball regions. The resolution of the

6

Page 7: ILASS Americas 28th Annual Conference on Liquid ... · 1.17 m with a 5x objective lens and a eld of view (FOV) of 2.25 1.4 mm. The smallest resolvable feature, found by imaging an

Holenumber

Hole inletdiameter (µm)

Hole length(µm)

Counterbore inletdiameter (µm)

Counterboredepth (µm)

L/D

1 175 154 380 397 0.882 173 172 380 398 0.993 173 168 381 397 0.974 173 161 386 396 0.935 172 143 384 396 0.836 172 139 384 386 0.817 171 139 383 391 0.818 171 139 383 396 0.81Mean 172 152 383 395 0.88Standarddeviation

1.3 13.8 2.1 4.1 0.08

Nominaldimensions

165 170 388 470 1.03

Percenterror (%)

4.55 10.66 1.39 16.04 14.55

Table 1: Table of a subset of critical nozzle dimensions. The hole inlet diameter refers to the diameterbetween the inlet and sac, and the counterbore inlet diameter refers the diameter between the hole andcounterbore. The percent error is between the target dimension and mean of the measured values.

reconstructed image set from 7–BM is coarsest atthe nozzle edges due to phase effects. We conserva-tively quote the resolution of these boundaries at 5µm, as the bright and dark fringes introduce ambi-guity into the edge detection. On the nozzle body,the diffraction–limited spatial resolution of 1.5 µmis expected to be coarsened slightly after the im-age set passes through a denoising filter, setting thelower bound on the resolution to 3 µm. Becausethe tomography setup has been tailored for imagingthrough dense material, high contrast between thenozzle and air is achieved. This distinct separationbetween materials is a prerequisite for effectively cre-ating an isosurface that preserves spatial resolution.

A threshold is set on the reconstructed imagestack in Seg3D using an Otsu method that segmentsthe intensity histogram into four collectively exhaus-tive bins. Manual clean–up of the threshold was re-quired in a thin band of poor contrast at the inter-face between the nozzle tip and relatively thicker in-jector body. The resulting isosurface was smoothedwith a Laplacian smoothing filter. Figure 7a showsa cut–away through the center of the isosurface ofthe GDI nozzle, exposing the inlet hole and coun-terbore structure, as well as the sac and check ballsurface finish. The clear distinction in surface rough-ness between the nozzle body and check ball lendsfavorable support that the surface finish on the noz-zle is to some degree resolved by the technique, i.e.there must be surface features on the nozzle body

that are as large or larger than the spatial resolu-tion of the reconstruction. Apart from the surfacefinish on the nozzle, the isosurface exposes inclusionsat the corner inlet on five of the eight nozzle holes.Figure 7b provides a closer look at two of the inclu-sions in the hole inlet corners. The average length ofthese inclusions is 19 µm, approximately 12% of thenominal hole diameter. These voids exist in crucialregions of the flow path, where cavitation is expectedat the inlet of the nozzle. As such, these inclusionsmay have an impact on the local flowfield, and cor-respondingly, the downstream fuel spray behavior.A summary of measurements of a subset of key noz-zle dimensions is given in Table 1 as compared tonominal dimensions for the Delphi GDI nozzle usedin this study. The hole orientation is defined bythe SAE J2715 standard for gasoline fuel injectors[20]. The tomographic reconstruction of the GDInozzle reveals hole–to–hole variation as well as di-vergence from the nominal geometry. The 9% vari-ation in the L/D ratio may have an impact on themass flow rate, velocity, spray angle, and breakuplength of each plume. Larger counterbore diametersmight also result in selectively increased recircula-tion inside the cavity, thus affecting the atomizationprocess. Indeed, counterbore diameter has been pre-viously linked to increased droplet size for any givenL/D ratio [21].

7

Page 8: ILASS Americas 28th Annual Conference on Liquid ... · 1.17 m with a 5x objective lens and a eld of view (FOV) of 2.25 1.4 mm. The smallest resolvable feature, found by imaging an

Summary and Conclusions

We have presented recent advancements in x–ray tomography measurements of injection nozzlesat the APS 7–BM beamline. An upgrade to the fa-cility has enabled x–rays at relatively higher photonenergy and flux, capable of partially transmittingthrough the steel body of an injection nozzle to ob-tain high contrast projections of its interior. Thespatial resolution of 1.5 µm is preserved as much aspossible throughout the imaging process by optimiz-ing the procedure to eliminate the need for softwareimage enhancement that generally relies on smooth-ing methods. These efforts include averaging a set ofimages at each sampled angle to increase SNR andimage contrast, while also providing multiple datapoints at each pixel location to temporally filter outresidual zingers. Proper shielding is used to limitzinger contamination, and the distance between thescintillator and nozzle is reduced to minimize phaseeffects along the nozzle edges. Lastly, a beam hard-ening correction is applied to the raw image set. Theabove steps result in high contrast, high resolutionimages that require minimal manual post–processingto isosurface. From image acquisition to isosurfaceoutput, the process requires less than two days andcan be performed on workstation–level equipmentexclusively using open–source software.

The reconstruction results of an eight–hole Del-phi GDI nozzle were presented as a demonstrationof the current capability. The isosurface reveals twodistinct surface finishes on the nozzle body and onthe check ball. In future work, the fidelity of thesurface roughness will be verified through scanningelectron microscope images and complementary to-mography measurements. Injectors manufacturedby various companies will be imaged in order tobetter understand the effect of varying manufactur-ing techniques on the near nozzle flowfields obtainedfrom x–ray radiography and tomography measure-ments. The reconstructed images provide measure-ments of the critical nozzle dimensions, revealinghole–to–hole variations as well as deviations from thetarget conditions. While variations in the hole cross–sectional area affect plume–to–plume mass flux dis-tributions, it has been shown that the variabilitydoes not explain the corresponding fuel spray distri-bution completely [6]. The effect of the L/D ratio,additional geometrical features, surface roughness,and plume–to–plume interactions all warrant furtherdeliberation to assess their individual and collectiveroles on the fuel spray behavior.

Acknowledgements

The research presented in this paper was per-formed at the 7–BM beamline at the Advanced Pho-ton Source at Argonne National Laboratory. Useof the APS is supported by the U.S. Departmentof Energy (DOE) under Contract No. DEAC02-06CH11357. Argonne’s x–ray fuel spray researchis sponsored by the DOE Vehicle Technologies Pro-gram under the direction of Gurpreet Singh and LeoBreton.

The authors would like to gratefully acknowl-edge Doga Gursoy, one of the developers of TomoPy,for useful discussion regarding the reconstruction al-gorithm. We acknowledge the use of computing re-sources provided on the Blues and Fusion high per-formance computing clusters operated by the Labo-ratory Computing Resource Center at Argonne Na-tional Laboratory.

References

[1] R. Payri, J.M. Garcıa, F.J. Salvador, and J. Gi-meno. Fuel, 84(551-561), 2005.

[2] R. Payri, F.J. Salvador, J. Gimeno, and L.D.Zapata. Fuel, 87(1165-1176), 2008.

[3] S. Som, A.I. Ramirez, D.E. Longman, and S.K.Aggarwal. Fuel, 90(1267-1276), 2011.

[4] A.L. Kastengren, F.Z. Tilocco, C.F. Powell,J. Manin, L.M. Pickett, R. Payri, and T. Bazyn.Atomization and Sprays, 22(12):1011–1052,2012.

[5] D.J. Duke, A.B. Swantek, N. Sovis, F.Z.Tilocco, C.F. Powell, A.L. Kastengren,D. Gursoy, and T. Bicer. SAE Int. J. Engines,2015-01-1873, 2016.

[6] P. Strek, D. Duke, A. Swantek, A. Kastengren,C.F. Powell, and D.P. Schmidt. SAE TechnicalPaper, 2016-01-0858, 2016.

[7] M. Zhang, M.C. Drake, and K. Peterson.ASME, ICEF2013-19066, 2013.

[8] R. Payri, J. Gimeno, P. Marti-Aldaravi, andD. Vaquerizo. SAE Technical Paper, 2015-01-1893, 2015.

[9] Jens Als-Nielsen and Des McMorrow. Elementsof Modern X-Ray Physics, chapter 9, pp. 307–313. John Wiley and Sons, 2011.

[10] Y. Wang, F. De Carlo, D.C. Mancini, I. Mc-Nulty, B. Tieman, J. Bresnahan, I. Foster, J. In-sley, P. Lane, G. von Laszewski, C. Kesselman,

8

Page 9: ILASS Americas 28th Annual Conference on Liquid ... · 1.17 m with a 5x objective lens and a eld of view (FOV) of 2.25 1.4 mm. The smallest resolvable feature, found by imaging an

MH. Su, and M. Thiebaux. Review of ScientificInstruments, 72(4):2062–2068, 2001.

[11] M. Pateyron, F. Peyrin, A. M. Laval-Jeantet,P. Spanne, P. Cloetens, and G. Peix. SPIE Med-ical Imaging, 2708(417-426), 1996.

[12] A. Kastengren, C.F. Powell, D. Arms, E.M.Dufresne, H. Gibson, and J. Wang. J Syn-chrotron Rad, 19:654–657, 2012.

[13] D. Gursoy, F. De Carlo, X. Xiao, and C. Jacob-sen. J Synchrotron Rad, 2014.

[14] Thorsten M. Buzug. Computed Tomography.Springer, 2008.

[15] B.A. Dowd, G.H. Campbell, and D.P. Siddons.Developments in synchrotron x-ray computedmicrotomography at the national synchrotronlight source. Technical report, Brookhaven Na-tional Laboratory, 1999.

[16] F. Marone and M. Stampanoni. J SynchrotronRad, 19:1029–1037, 2012.

[17] A. Chambolle. Journal of Mathematical Imag-ing and Vision, 20(89-97), 2004.

[18] CIBC. Seg3d, 2015.

[19] SPIE Proceedings. Status of XOP: v2.4: re-cent developments of the x-ray optics softwaretoolkit, volume 8141, 2011.

[20] Gasoline fuel injector spray measurement andcharacterization. SAE International, 2011.

[21] P. Tu, H. Xu, D. Srivastava, K. Dean, D. Jing,L. Cao, A. Weall, and J.K. Venus. SAE Tech-nical Paper, 2015-01-1906, 2015.

9