14th International Symposium on Particle Image Velocimetry – ISPIV 2021
August 1–5, 2021

Simultaneous Particle Tracking and Temperature
Measurements during Additive Manufacturing using a

High-speed Spectral Plenoptic Camera
Dustin Kelly1, Ralf Fischer2, Ari Goldman1, Sarah Morris1, Bart Prorok2, and

Brian Thurow1

1 Auburn University, Advanced Flow Diagnostics Laboratory, Auburn, AL USA
2 Auburn University, Materials Research and Education Center, Auburn, AL USA

In this work, a high-speed spectral plenoptic camera was used for three-dimensional (3D) simultaneous
particle tracking and pyrometry measurements of hot spatter particles ejected during the metal additive man-
ufacturing process. Additive manufacturing (AM) has an increasing role in the aerospace, energy, medical
and automotive industry (DebRoy et al., 2018). While this new technology enables the production of highly
advanced parts, research on the fundamental mechanisms governing the laser-matter interactions are an on-
going challenge because of the spatial and temporal resolution inherent to the AM process. One challenge
is the characterization of spatter particles ejected from the melt pool, as these particles can be incorpo-
rated into the final part affecting the mechanical properties (Deng et al., 2020). One potential solution for
simultaneously measuring velocity and temperature of the spatter particles is the spectral plenoptic camera.

Figure 1: (Left) A simple schematic of a three color spectral plenoptic camera showing how a single
microlens captures an image of the aperture plan, and (right) the band pass filter array layout for these
experiments where 800 nm is red, 600 nm is orange, and 532 nm is green.

The plenoptic camera is a type of light-field camera, which provides the ability to instantaneously cap-
ture both angular and spatial information with one sensor. Adelson and Wang (1992) developed the modern
design of a plenoptic camera, which is composed of a microlens array placed in front of a relay lens that
images the microlens array onto a traditional sensor. Each microlens of the array images the aperture plane,
where each pixel discretizes the aperture, capturing the angular information as shown in Figure 1. A conve-
nient way to parameterize plenoptic images is through two-plane parameterization as described by Levoy
(2006), where the u-v and s-t planes correspond to the aperture and microlens planes, respectively. Plenoptic
imaging provides the ability to generate perspective images of the scene by adjusting the (u,v) coordinate.
A recent modification of the plenoptic camera, first introduced by Danehy et al. (2017), adds spectral filters
at the aperture plane of the main lens. Each microlens captures sub-images of the spectral filters as shown in
Figure 1. The spectral plenoptic camera provides the ability to individually select spectral filters, where each
filter possesses limited angular information. Increasing the number of different spectral filters decreases the
overall angular information for a particular wavelength/spectrum.

The current implementation of the spectral module and the high-speed plenoptic camera has a maximum
of seven filters as shown in Figure 1, in which there were three different spectral 10 nm band-pass filters as
indicated by the filter layout image. The camera used was a Phantom VEO4k high-speed camera capable of
1000 Hz at a resolution of (2048x2048) with a microlens array (471x362) placed in front of a relay lens that
relays the microlens images to the sensor. The spectral module is composed of two lenses designed for a
working distance/magnification and to f-match the microlens array as described by Fahringer et al. (2018),
where a magnification of 0.6 was chosen for these experiments.



The first step of the PTV algorithm is locating the particles in the volume for each frame using plenoptic
ray bundling, first described by Clifford et al. (2019). Ray bundling takes advantage of the two plane param-
eterization, where perspective images are first generated and particle locations are found through applying
a minimum threshold. The perspective is the (u,v) coordinate and the location on the image is the (s,t)
coordinate of the rays found during thresholding. Every ray is then traced between the two planes, where
the particle locations correspond to the point where the rays of different perspectives converge. This corre-
sponds to a 3D location in the image space, which is converted to the object space. Velocity measurements
were made using a central finite time difference particle tracking technique, wherein particles’ 3D locations
were matched in the images. Ray bundling was conducted with the (i) 600 nm filter perspectives, (ii) 800
nm filter perspectives, and (iii) all perspectives to determine the uncertainty.

Dual-wavelength pyrometry was used to estimate the particle temperatures, where temperature is de-
termined by the ratio of the average intensity between 600 nm and 800 nm filters for each microlens
(r = I600nm/I800nm). Temperature measurements are limited to the focal plane of the camera due to the
assumption that the spectral information captured by one microlens originates from one particle, which only
occurs at the focal plane of the camera. Further improvements to the pyrometry algorithm could remove this
limitation. The temperature calibration was performed in a Lindberg/Blue M tube furnace. An Inconel 625
shim with a thermocouple attached to it was placed in the center of the furnace and temperature calibration
images were captured at temperatures ranging between 950◦C and 1250◦C at intervals of 50◦C.

The AM experiment was performed on a customized AM machine at Auburn University’s Materials
Research and Education Center. The laser source was an IPG YLR-400AC continuous wave ytterbium fiber
laser with a wavelength of 1070 nm and a maximum output power of 400 W. An IPG Mid-Power Scanner
with a 250 mm f-theta lens was used to deliver the beam onto a 10x10 mm Inconel 625 substrate. An Inconel
625 powder layer of about 30 µm was spread onto the substrate. The laser scan path was a single continuous
line scan with 800 mm/s scan velocity and 250 W laser power, and the laser traversed the z=0 plane of
the volume from right to left in Figure 2. The 8.5 mm long scan was performed below the scan head and
perpendicular to the direction of view of the plenoptic camera. The single line scan was performed under
ambient conditions for the demonstration of the capabilities of the plenoptic camera.

Figure 2: (a) Spectral plenoptic image with a zoomed image with a blue circle corresponding to a single
microlens that images a particle, (b) particle locations with the color corresponding to temperature with
velocity vectors (mm/ms).

Figure 2(a) shows a spectral plenoptic image of a high temperature particle field generated by the ad-
ditive manufacturing process. Figure 2(b) shows particle locations along the x-y plane calculated using ray
bundling where the particle color corresponds to the temperature. The velocity vector size is scaled with
velocity (mm/ms). Particle velocities ranged between 0.3 m/s to 7 m/s, where the faster particles originated
in the first 5 ms, which agrees with the velocities found by Guo et al. (2018). Traditionally with single
plenoptic reconstructions, the uncertainty in the z-direction is 6 to 10 times larger than in the x and y direc-
tions. This is on a similar order even with the reduced angular information per wavelength. The majority
of particle temperatures ranged between 1000◦C to 1600◦C. With a melting temperature of approximately
1300◦C, the Inconel 625 particles were present in both liquid and solid states. For particles along the focal
plane, a general cooling trend of approximately 150◦C/ms was observed. The measurement uncertainty for
both particle location and temperature are still a current topic of investigation.



Acknowledgements
This work was sponsored by the United States National Institute of Standards and Technology under con-
tracts NIST-70NANB16H272 and NIST-70NANB18H220. The authors gratefully acknowledge the support
of Mr. Steven Moore and Mr. Emre Kayali at Auburn University’s Materials Research and Education Center.

References
Adelson E and Wang J (1992) Single lens stereo with a plenoptic camera. IEEE Transactions on Pattern

Analysis and Machine Intelligence 14:99–106

Clifford C, Tan Z, Hall E, and Thurow B (2019) Particle matching and triangulation using light-field ray
bundling. 13th International Symposium on Particle Image Velocimetry

Danehy PM, Hutchins WD, Fahringer TW, and Thurow BS (2017) A Plenoptic Multi-Color Imaging Py-
rometer. 55th AIAA Aerospace Sciences Meeting pages 1–7

DebRoy T, Wei H, Zuback J, Mukherjee T, Elmer J, Milewski J, Beese A, Wilson-Heid A, De A, and Zhang
W (2018) Additive manufacturing of metallic components – process, structure and properties. Progress
in Materials Science 92:112–224

Deng P, Karadge M, Rebak RB, Gupta VK, Prorok BC, and Lou X (2020) The origin and formation of
oxygen inclusions in austenitic stainless steels manufactured by laser powder bed fusion. Additive Manu-
facturing 35:101334

Fahringer TW, Danehy PM, and Hutchins WD (2018) Design of a Multi-Color Plenoptic Camera for Snap-
shot Hyperspectral Imaging. 2018 Aerodynamic Measurement Technology and Ground Testing Confer-
ence pages 1–9

Guo Q, Zhao C, Escano LI, Young Z, Xiong L, Fezzaa K, Everhart W, Brown B, Sun T, Chen L, and
et al (2018) Transient dynamics of powder spattering in laser powder bed fusion additive manufacturing
process revealed by in-situ high-speed high-energy x-ray imaging. Acta Materialia 151:169–180

Levoy M (2006) Light Fields and Computational Imaging. Computer 39:46–55