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

Applications of particle tracking velocimetry to severe
nuclear accident experimentation

M. Johnson1, C. Journeau1∗
1 CEA, DES, IRESNE, DTN, SMTA, LEAG, Cadarache, 13115 Saint-Paul-lez-Durance, France

∗ christophe.journeau@cea.fr

Abstract
Experimental research into severe nuclear accidents may entail the discharge of a very high-temperature
lava-like molten fuel mixture, corium, either into a pool of less-dense, more-volatile coolant or onto a solid
substrate where the corium will spread and cool. In both instances, remote, high-speed video imaging is
usually required to interpret these transient interactions and PTV represents a powerful tool for the charac-
terisation of the dynamic properties of discrete melt fragments or distinctive features in the surface of the
melt during spreading. Nuclear fuel-coolant interactions present particular challenges for PTV analysis as
a molten jet and its fragments can exhibit high rates of inter-frame deformation and undergo fragmentation
with a relatively high frequency. A PTV algorithm, adapted to these challenges, is presented whereby a
user-defined tolerance in the evolution of certain particle properties is used to refine the potential candidate
particles prior to particle matching. This candidate refinement step is used to distinguish between accept-
able levels of deformation between successive sightings of a given particle, and more substantial changes
consistent with fragmentation or coalescence, requiring the tracking of a new particle. Implementation of
the PTV algorithm is presented for (1) an X-ray video from the FCINA-30-1 experiment between a jet of
molten stainless steel and liquid sodium, conducted at the JAEA’s MELT facility, and (2) video imaging of
the VE-U9-ceramic experiment of a molten corium-thermite mixture spreading on a zirconium substrate,
conducted at the CEA’s VULCANO facility. The latter case-study enabled the characterization of > 70,000
local velocity vectors at locations corresponding to distinctive temperature heterogeneities in the surface of
the spreading melt, providing extensive insight into the spreading dynamics for the validation of corium
spreading models.

1 Introduction
During a hypothetical core disruptive accident (CDA) in a nuclear reactor, nuclear fuel and core infras-
tructure would rapidly melt leading to the relocation of molten core material, a high-temperature lava-like
substance referred to as corium. One possible progression of the CDA concerns the discharge of coherent
jets of molten nuclear fuel through core discharge tubes, designed to reduce reactivity in the core (Bertrand
et al. (2018)), or control rod guide tubes, to a pool of the less dense, more volatile coolant, leading to a fuel-
coolant interaction (FCI). This phase of the CDA is of interest to nuclear engineers as the fine fragmentation
of the molten fuel, sudden destabilisation of vapour films surrounding melt fragments and rapid heat transfer
between melt and coolant raises the potential for vapour explosions (Berthoud (2000)).

A subsequent phenomenon of interest concerns the possibility that the reactor pressure vessel is breached
by the high-temperature corium, resulting in corium discharge to the containment floor (Dinh et al. (2000);
Journeau et al. (2003)). Under these circumstances it is desirable to promote ex-vessel spreading over a
broad surface area to enhance the dissipation of radioactive decay heat and preserve the integrity of the
containment building (Johnson et al. (2021a)). Both of these hypothetical scenarios have been the subject
of extensive experimental research. The JAEA’s MELT facility (Matsuba et al. (2016); Emura et al. (2019);
Johnson et al. (2021b)) and CEA’s KROTOS facility (Zabiego et al. (2010); Tyrpekl et al. (2014)) have been
used to investigate FCIs for sodium-cooled and water-cooled reactor configurations respectively, using X-ray
imaging to visualise the interactions within the optically opaque test vessels (Berge et al. (2018); Johnson
et al. (2021b)). The CEA’s VULCANO facility has been employed to investigate the discharge of molten
corium-thermite mixtures and their subsequent spreading on concrete or ceramic substrates (Journeau et al.



(2003, 2006); Johnson et al. (2021a)). Mutliple optical and infra-red cameras are used to observe the pro-
gression of the melt during spreading.

Particle tracking velocimetry (PTV) (Crocker and Grier (1996); Brevis et al. (2011); Ohmi and Li (2000))
is a powerful tool for the interpretation of images acquired during FCI and corium spreading experiments.
Particle-matching enables analysis of the population of melt fragments without the double counting bias of
the multiple sightings of each particle. The estimated velocity of missile fragments and the rate of expansion
of vapour clouds could also be used to infer the energetics of fragmentation events. The spreading velocity of
an advancing melt front and local velocity distribution similarly provide crucial insight into melt spreading
dynamics for the validation of severe accident codes (Spindler and Veteau (2006); Wittmaack (2002)).

Each type of experiment presents unique challenges for PTV analysis. The hydrodynamic pre-mixing
of a molten metallic jet in a less dense fluid represents a complex three-dimensional transient interaction,
captured by a single two-dimensional projection of the attenuated fan-beam X-ray source. The interac-
tion occurs within a timescale of a few seconds, during which the particle population evolves significantly.
A fragmentation event generates multiple potential candidate particles in close proximity, the new trajec-
tories of which may either be tracked as a continuation of the original particle or characterized as new
particles depending on the rigidity particle-matching criteria. Coalescence events present similar complica-
tions, particularly as these can be difficult to distinguish from quasi-coalescence whereby distinct particles
temporarily by-pass each other at different depths in the X-ray photon path. Different FCI experimental
facilities generate X-ray images with frame rates ranging from 100–1000 fps (Zabiego et al. (2010); Emura
et al. (2019); Johnson et al. (2021b)) and the size of imaging fields of view vary significantly, requiring that
the PTV algorithm tolerates a wide range of particle densities and offers flexibility in the particle matching
criteria to account for the different rates of inter-frame particle deformation. A PTV algorithm is presented
whereby candidate particle pairings are assessed in terms of a weighted similarity index enabling the re-
jection of candidate particles of insufficient likeness to the paired sighting prior to particle matching. The
application of this PTV approach is demonstrated for (1) the FCINA-30-1 FCI experiment (Johnson et al.
(2021b)) between a jet of molten steel and liquid sodium, performed at the MELT facility, and (2) the
VE-U9-ceramic experiment (Johnson et al. (2021a)) of a corium-thermite melt spreading on a zirconium
substrate, performed at the VULCANO facility.

2 Experimental methods

2.1 Fuel-coolant interaction experiments
FCI experimental facilities, such as KROTOS and MELT, consist, in simplest terms, of a crucible for the
melting of metal and metal oxide nuclear fuel analogues, a funnel to discharge the melt as a coherent jet, a
test section containing a pool of liquid coolant, and a radiographic imaging system to visualise the interac-
tion within the optically opaque environment. To demonstrate the application of PTV to FCI experimental
images, the FCINA-30-1 experiment (Johnson et al. (2021b)) conducted at the JAEA’s MELT facility (Mat-
suba et al. (2016); Emura et al. (2019)) will be investigated. This experiment was conducted between a jet
of molten stainless steel and a pool of sodium coolant. A detailed explanation of the MELT facility and
experimental procedure can be found in Johnson et al. (2021b). The radiographic images were acquired at
a frame rate of 1000 fps for a field of view in the order of ∼ 12cm in diameter (Johnson et al. (2021b)).
This FCI research at the MELT facility was performed under the framework of the current implementing
arrangement of the France-Japan collaboration on Sodium-cooled Fast Reactors (SFRs) from 2020 to 2024.

2.2 Ex-vessel corium spreading experiments
The VULCANO facility consists of a zirconium crucible with a discharge tube at its base, suspended above
a spreading test section with boundary walls which diverge at an angle of 9.5°. The VE-U9 spreading
tests discharge a ≈ 30kg molten corium-thermite mixture to spreading test sections with either a ceramic
zirconium or sacrificial concrete substrate. The spreading is imaged from multiple perspectives using optical
and infrared cameras stored in protective housings. Complete details of the experimental procedure are
presented in Johnson et al. (2021a).



Figure 1: Flow chart for the PTV algorithm.

2.3 Image processing and analysis
The images generated during FCI and ex-vessel spreading experiments are processed and subsequently
analysed using SPECTRA (Software for Phase Extraction and Corium TRacking Analysis), a CEA image
analysis software developed using Matlab 2021a (Mathworks, USA) (Johnson et al. (2021b)). SPECTRA
is used to (1) clean the images of artefacts, such as additive white noise, lens distortion and vignetting, (2)
extract objects of interest for analysis, and (3) perform quantitative characterization of the detected objects,
including the implementation of tracking by PTV.

2.3.1 Particle tracking velocimetry

The PTV framework implemented in SPECTRA, presented as a flow chart in Figure 1, is based on the
algorithm proposed in Crocker and Grier (1996) for images of colloidal particles but uses an alternative
approach for particle matching more specific to the requirements of FCI experimental images. The objective
of the alternative particle matching routine is to allow the user to define bespoke, video-specific criteria to
distinguish minor particle deformations from more significant fragmentation or coalescence events which
constitute the formation of a new particle.

Following image pre-processing and segmentation, all detected particle sightings are identified, labelled
and characterized according to K digital particle properties, not limited to the centroid, area, perimeter,
caliper dimensions, orientation, mean and maximum intensity, aspect ratio and circularity. The PTV algo-
rithm proceeds with the pre-selection of candidate particles, which is performed by distance thresholding.
The centroid coordinates of j candidate particles in frame i+ 1 (xi+1, j) will have an Euclidean distance
from a particle centroid in frame i of less than a defined maximum displacement, R, (Ohmi and Li (2000))
therefore satisfying the inequality:

|xi+1, j− xi| ≤ R (1)

Candidate particle assessment is performed using the K properties of each digital particle. The corre-
sponding kth property of the current frame particle and the j candidate particles are labelled yi,k and yi+1, j,k
respectively. Each property is assigned a weighting coefficient, ωk, to refine the array of particle properties
to a single coefficient called a weighted similarity index, WSI, calculated according to eq. 2:

WSI =
1

∑
K
k=1 ωk

K

∑
k=1

ωk
|yi,k− yi+1, j,k|

yi,k
≤MWSI (2)

The WSI diminishes with increased likeness between the current frame and candidate particle. If only
a single property, such as the area, is assigned a non-zero weighting coefficient, the WSI represents the
percentage change in that property between consecutive video frames, normalised by a factor of 100. The
candidate particles are refined by eliminating those with a WSI greater than a threshold maximum weighted
similarity index, MWSI. The MWSI defines an acceptable tolerance in the evolution in particle properties
between consecutive frames. This MWSI coefficient should be set to allow for sufficient tolerance in both
the real particle deformation between frames and the more artificial uncertainties in particle detection and
characterization due to imaging artefacts. The candidate particle assessment and refinement steps can be
performed iteratively using different ωk and MWSI values to eliminate candidate particles which fail to
satisfy alternative similarity criteria.

The remaining candidate particles, which have satisfied the requisite conditions of proximity and phys-
ical likeness, proceed to the particle matching stage. The current frame particle and candidate particle pair



(a) (b) (c)

Figure 2: (a) An example X-ray image from the FCINA-30-1 experiment (reproduced from Johnson et al.
(2021b)), (b) the same image following pre-processing and melt segmentation, cropped to a region of inter-
est, and (c) a semi-transparent overlay of (b) and the subsequent frame with the centroids indicated as red
circles (current frame) and blue crosses (subsequent frame).

with the lowest WSI are assigned the same particle number and are subsequently removed from alternative
candidate particle pairings. The process is repeated until no valid particle pairs remain within the two im-
ages under consideration. The algorithm repeats this process for the subsequent image pairs. On completion
of the particle matching and labelling process, unlabelled particles and particles which present with fewer
than the desired minimum sightings, are removed. Particles are then renumbered sequentially, if preceding
particle numbers have been eliminated, and the characterization data is sorted by ascending particle number
and then by frame number.

The final stage requires the estimation of velocity vectors. In most cases the velocity vectors are deter-
mined from the Euclidean displacement of the centroid, or intensity weighted centroid, normalised by the
frame interval. For objects cropped at the image border, such as a molten jet entering the imaging window
from above, the centroid displacement does not provide an accurate measure of the trajectory. For these
border objects, the compass points of a rectangular bounding box around the object are defined and the dis-
placement of the coordinates opposite the cropped boundary (i.e. the southern coordinate of the bounding
box for objects entering from the top of the image) are used to estimate the particle trajectory.

3 Results and discussion

3.1 Fuel-coolant interactions
An example X-ray image from the FCINA-30-1 experiment, reproduced from Johnson et al. (2021b), is pre-
sented in Figure 2a. The image suffers from some common X-ray imaging artefacts, such as loss of intensity
towards the image border (vignetting) and considerable additive white noise. The images are denoised in
two stages using a spatiotemporal video-block matching and 3D filter, V-BM3D, (Danielyan et al. (2012)),
followed by a non-local means, NLM, (Buades et al. (2005)) filter to remove residual noise. The images are
then normalised by a clean flat field image, acquired from the averaged frames prior to the arrival of the melt
jet, and corrected for pincushion distortion. The pre-processed images are then dynamically thresholded and
cropped to a region of interest below the initial surface of the sodium pool and within an internal cylinder
within the test section. Application of this image pre-processing and segmentation routine to the image in
Figure 2a generates the image in Figure 2b. A compound image of the melt fragments in Figure 2b with a
semi-transparent overlay of those observed in the subsequent frame, with their centroids shown as red circles
and blue plus signs respectively, is presented in Figure 2c.

These image data have a number of advantageous qualities for exploitation by PTV. The images are



(a) (b)

Figure 3: (a) The segmented image from Figure 2b overlaid with velocity vectors revealing the complete
trajectories of each particle, and (b) a chronological montage of all sightings of the coherent jet (particle 79)
prior to its fragmentation.

high in resolution giving a relatively low particle density of ∼ 10−5 pixel−1. The particles are macroscopic,
morphologically distinctive, and are often relatively simple to distinguish. The high frame rate dictates a
relatively low inter-frame particle displacement, often significantly smaller than the particle dimensions, and
so particle pairings are typically self-evident. The complexity in applying PTV arises from (1) high rates
of particle deformation due their molten nature, (2) a high frequency of particle division and coalescence,
(3) apparent or quasi-coalescence due to the flattening of the depth field in the 2D image projection, (4) the
partial or total obscuration of some particles by vapour clouds, (5) the possible detection of false particles
due to the significant additive white noise in the raw images, and (6) the fact that the interaction extends
beyond the limits of the field of view, thereby cropping many sightings at the image border. Figure 2c
identifies seven distinct first frame melt sightings, with eight sightings observed in the subsequent frame. Of
the seven first frame particle sightings, two are cropped at the image border, entering and leaving the imaging
window respectively, four sightings appear to coalesce to form two particles in the subsequent frame, one
particle fragments into two daughter particles of similar dimensions, and two sightings in the second frame
appear to be false particles detected due to the image noise.

For these X-ray projections of melt fragments, the intensity profile of the particle is proportional to the
particle dimension, or chord length, parallel to the photon flux (Macovski (1983)), thus, small particles may
present with a very low signal to noise ratio. Despite the implementation of powerful image denoising fil-
ters, the detection of sub-millimetre particles will generally necessitate the detection of some false particles,
corresponding to small pockets of residual noise. Since these false particles present at random spatiotempo-
ral locations, rather than in close proximity to previous sightings, an initial iteration of the PTV algorithm
with a conservative (relatively large) maximum displacement and a minimum sightings threshold of at least
2 will eliminate the majority of false particles. These false particles can be effaced from the video in a
PTV-segmentation step, prior to a second, more precise iteration of the PTV algorithm with more stringent
tracking parameters. Coupling the conventional intensity and size based thresholding of image segmenta-
tion with the distance and similarity thresholding of this PTV-segmentation step to eliminate false particles
can thereby enable particle detection at a reduced signal to noise ratio. Implementation of the PTV algo-
rithm with a maximum displacement of 50 pixel and a minimum sightings threshold of 2, with no similarity
threshold (MWSI→ ∞), eliminates 1398 false particles from the population of 10422.

Refining the tracking parameters to a maximum displacement of 35 pixel, allowing for a 20 % tolerance
in the change in particle area between successive frames (this tolerance is relaxed for objects entering and



leaving the imaging window) and a minimum sightings threshold of 5 identifies 338 particles with 4547
sightings. Thus, roughly half the total number of objects detected in the video correspond to long-life
particles presenting with, on average, 13 sightings per particle. All of the velocity vectors corresponding
to the 7 particles observed in Figure 2b, and one additional particle (particle 106) produced following a
quasi-coalescence involving particle 76, are presented, overlaid on the segmented image, in Figure 3a. The
tracking history of particles 68 and 76 reveals a shared origin which implies particle 68 to be a daughter
particle following the fragmentation of particle 76. The most interesting observation from Figure 3a con-
cerns the sudden reversal in the trajectory of particle 106 close to the bottom left corner of the image. Closer
inspection of the video revealed that the particle in question experiences an elastic collision with a thermo-
couple installed on the steel support rod observed in the left hand side of Figure 2a. The elastic collision not
only infers solid-like behaviour, confirming the formation of a solid outer crust at the melt-coolant interface,
but indicates that within 90 ms of contact with the coolant, the melt crust grows to a sufficient thickness to
withstand a ≈0.9 ms−1 collision with the thermocouple without fracturing.

Figure 3b presents a chronological montage of all sightings of particle 79, the coherent melt jet, prior to
its fracture. The morphology of the jet similarly exhibits solid-like qualities within ≈ 20ms of contact with
the sodium. After around 40 ms of entering the imaging window, the jet spans the height of the imaging
window. While the PTV algorithm continues to recognise the image spanning jet as the same particle,
the characterization of the velocity vectors ceases to function for the image-spanning sightings since the
centroid, intensity weighted centroid and the coordinates of the bounding box all become misleadingly
static and fail to accurately capture the inter-frame displacement. For objects which span the length of the
imaging window an alternative method will be required for the spatial cross-correlation of adjacent sightings
to determine the particle trajectory. A normalized 2D cross-correlation (Lewis (1995)) approach is currently
being implemented in SPECTRA for the determination of the trajectories of these selected image spanning
particles.

3.2 Ex-vessel spreading
Figure 4a presents an example image from the VE-U9-ceramic melt spreading experiment (Johnson et al.
(2021a)). The melt is segmented from each image by dynamic thresholding and a mask of the spreading
section floor is used to refine the melt to its footprint. Corners of the squares in the checkerboard image
in Figure 4b are defined as control points, used to generate a transform function to correct the camera per-
spective to a bird’s-eye view of the melt. The product of these segmentation and transformation steps is
presented in Figure 4c.

Application of PTV to the melt footprint in Figure 4c enables characterization of the velocity of the
advancing melt front as presented in Johnson et al. (2021a), however this provides only limited insight
into the spreading dynamics. The melt surface appears highly heterogeneous in temperature, providing a
distinctive pattern in the melt luminosity. Convolution of the image with an 11× 11 pixel Laplacian of
Gaussian (LoG) operator with a standard deviation of 2, highlights the features with a significant change
in local luminosity, corresponding mainly to the outer melt contour and a series of local heterogeneities in
the melt surface temperature, generally features of reduced luminosity implying cold-spots, as presented in
Figure 4d. Size thresholding of the features to retain those comprised of 5-50 pixels isolates the localised
cold-spots from the larger features. PTV is employed to track these surface features and reveal the evolution
of the local surface velocity distribution with time and space.

Application of the LoG edge detection and size thresholding to the VE-U9-ceramic experimental images
revealed a total of 96755 surface features. Modeling of the VE-U9 spreading dynamics in Johnson et al.
(2021c) predicted surface velocities of up to around 0.25 ms−1. The maximum velocity, combined with a
pixel resolution in the transformed image of around 1.6 mm and a frame rate of 50 fps, dictates that local
surface features are unlikely to move more than 4 pixel between consecutive frames. This implies that many
features will exhibit displacements of less than a pixel and so velocity vectors will need to be averaged over
multiple frames for accurate representation.

The PTV algorithm was employed with a maximum displacement of 6 pixel, allowing for a 20 % change
in the area and mean intensity of features between consecutive frames, and a minimum sightings threshold
of 5. A total of 71527 sightings were found for 4534 distinct features, enabling the characterization of
local velocity vectors at 74 % of the locations where surface temperature heterogeneities were detected.
The instantaneous local velocities are averaged over five frames (or 0.1 s intervals) to smooth the data and to
detect sub-pixel displacements. The local velocities are presented for the frames at 2 and 5.7 s after the initial
melt discharge to the spreading section in Figures 4e and f respectively. While a maximum velocity, parallel
to the principal flow direction, vx, was found to be 0.45 ms−1, 98 % of the velocity vectors in the principal



(a) (b)

(c)

(d)

0.1 m/s

(e)

0.1 m/s

(f)

Figure 4: Images of (a) corium-thermite spreading on a zirconium substrate during the VE-U9-ceramic test,
(b) a checkerboard calibration image of 5× 5cm squares, (c) the segmented melt from (a), transformed
to a birds-eye perspective, (d) a Laplacian of Gaussian transform of (c) revealing the melt contour and
temperature heterogeneities in the melt surface, and (e-f) quiver plots of the PTV-determined velocity vectors
after (e) 2 s and (f) 5.7 s; the images in (c-f) correspond to dimensions of 830×340mm.

flow direction were less than 0.23 ms−1, which demonstrates good agreement with theoretical predictions
from Johnson et al. (2021c).

4 Conclusions
Experimentation in the field of severe nuclear accident research often requires remote video imaging of high-
temperature transient interactions. PTV represents a very powerful tool for the analysis of (1) radiographic
images acquired during investigations of molten fuel-coolant interactions, and (2) optical and infrared im-
ages acquired of ex-vessel melt spreading. Fuel-coolant interactions present particular challenges for PTV
analysis as the molten fuel exhibits high rates of deformation and a high frequency of fragmentation. A
bespoke PTV algorithm is presented whereby candidate particles are refined by user-defined similarity cri-
teria prior to particle matching, enabling the distinction between the tolerable deformation of a molten fuel
droplet and a more significant fragmentation event which constitutes the formation of new particles.

Application of the PTV algorithm to the FCINA-30-1 fuel-coolant interaction experiment (Johnson et al.
(2021b)) between molten stainless steel and liquid sodium enabled the detection of 338 relatively long-life
particles presenting with an average of 13 sightings each. One melt particle presented with a sudden reversal
in its trajectory which revealed an elastic collision with a thermocouple installed within the test section. This
distinctly solid-like behavior confirms that within 90 ms of first contact with the liquid sodium, the steel
formed a solid crust at the melt-coolant interface of sufficient thickness to withstand a 0.9 ms−1 collision
with the thermocouple, implying a crude lower limit for the rate of crust growth. Certain of the tracked
particles, representing large melt jet fragments with strong evidence of a frozen outer shell, spanned the
height of the imaging window and inhibited the characterisation of velocity through the displacement of the
centroid or the leading front. The PTV algorithm will require an alternative cross-correlation technique to
estimate the motion of objects with dimensions greater than the field of view. A second application of the
PTV algorithm to images acquired during the VE-U9-ceramic spreading test (Johnson et al. (2021a)) enabled



the tracking of distinctive surface temperature heterogeneities, or cold-spots, in the molten corium-thermite
mixture during its spreading on a ceramic substrate. This enabled the characterisation of over 70,000 local
velocities during the ≈ 10s spreading test, providing extensive data for the validation of ex-vessel corium
spreading simulations.

Acknowledgements
The authors would like to thank the JAEA, in particular everybody from the MELT team, for sharing their
FCINA-30-1 experimental data under the France-Japan collaboration on the SFR severe accident studies.
We also wish to thank Mitsubishi Heavy Industries for their financing of the VE-U9 spreading tests in
collaboration with the CEA, and the CEA’s PLINIUS team for their management of the VULCANO facility
during the VE-U9 tests.

References
Berge L, Estre N, Tisseur D, Payan E, Eck D, Bouyer V, Cassiaut-Louis N, Journeau C, Tellier RL, Singh

S, and Pluyette E (2018) Fast high-energy X-ray imaging for severe accidents experiments on the future
PLINIUS-2 platform. IEEE Transactions on Nuclear Science 65:2573–2581

Berthoud G (2000) Vapor explosions. Annual Review of Fluid Mechanics 32:573–611

Bertrand F, Marie N, Bachrata A, Brun-Magaud V, Droin J, Manchon X, Herbreteau K, Farges B, Carluec
B, Poumerouly S, and Lemasson D (2018) Status of severe accident studies at the end of the conceptual
design of ASTRID: Feedback on mitigation features. Nuclear Engineering and Design 326:55–64

Brevis W, Niño Y, and Jirka GH (2011) Integrating cross-correlation and relaxation algorithms for particle
tracking velocimetry. Experiments in Fluids 50:135–147

Buades A, Coll B, and Morel JM (2005) A review of image denoising algorithms, with a new one. Multiscale
Modeling Simulation 4:490–530

Crocker JC and Grier DG (1996) Methods of digital video microscopy for colloidal studies. Journal of
Colloid and Interface Science 179:298–310

Danielyan A, Katkovnik V, and Egiazarian K (2012) BM3D frames and variational image deblurring. IEEE
Transactions on Image Processing 21:1715–1728

Dinh TN, Konovalikhin MJ, and Sehgal BR (2000) Core melt spreading on a reactor containment floor.
Progress in Nuclear Energy 36:405–468

Emura Y, Isozaki M, Matsuba K, and Kamiyama K (2019) Penetration behavior of molten stainless steel
into a sodium pool. in Proceedings of the 24th national symposium on power and energy systems (SPES
2019). page 583

Johnson M, Denoix A, Bouyer V, Goda H, Kamohara SS, Takeuchi J, Brissonneau L, and Journeau C
(2021a) High-temperature ex-vessel corium spreading. part 1: Experimental investigations on ceramic
and sacrificial concrete substrates. submitted to Journal of Nuclear Science and Technology

Johnson M, Journeau C, Matsuba K, Emura Y, and Kamiyama K (2021b) Characterization of high-
temperature nuclear fuel-coolant interactions through X-ray visualization and image processing. Annals
of Nuclear Energy 151:107881

Johnson M, Schiano T, Denoix A, Bouyer V, and Journeau C (2021c) High-temperature ex-vessel corium
spreading. part 2: Scaling principles for gravity-viscous spreading with slip at the melt-substrate interface.
submitted to Journal of Nuclear Science and Technology

Journeau C, Boccaccio E, Brayer C, Cognet G, françois Haquet J, Jégou C, Piluso P, and Monerris J (2003)
Ex-vessel corium spreading : results from the vulcano spreading tests. Nuclear Engineering and Design
223:75–102



Journeau C, Haquet JF, Spindler B, Spengler C, and Foit J (2006) The VULCANO VE-U7 corium spreading
benchmark. Progress in Nuclear Energy 48:215–234

Lewis JP (1995) Fast template matching. in Vision Interface. pages 120–123

Macovski A (1983) Medical imaging systems. Prentice-Hall. 1st edition

Matsuba KI, Isozaki M, Kamiyama K, and Tobita Y (2016) Distance for fragmentation of a simulated
molten-core material discharged into a sodium pool. Journal of Nuclear Science and Technology 53:707–
712

Ohmi K and Li HY (2000) Particle-tracking velocimetry with new algorithms. Measurement Science and
Technology 11:603–616

Spindler B and Veteau JM (2006) The simulation of melt spreading with thema code part 1 : Model as-
sessment strategy and assessment against analytical and numerical solutions. Nuclear Engineering and
Design 236:415–424

Tyrpekl V, Piluso P, Bakardjieva S, and Dugne O (2014) Material effect in the nuclear fuel-coolant inter-
action: Analyses of prototypic melt fragmentation and solidification in the KROTOS facility. Nuclear
Technology 186:229–240

Wittmaack R (2002) Simulation of free-surface flows with heat transfer and phase transitions and application
to corium spreading in the EPR. Nuclear Technology 137:194–212

Zabiego M, Brayer C, Grishchenko D, baptiste Dajon J, Fouquart P, Bullado Y, Compagnon F, Correggio P,
françois Haquet J, and Piluso P (2010) The KROTOS KFC and SERENA/KS1 tests: experimental results
and MC3D calculations. pages 1–11


	Introduction
	Experimental methods
	Fuel-coolant interaction experiments
	Ex-vessel corium spreading experiments
	Image processing and analysis
	Particle tracking velocimetry


	Results and discussion
	Fuel-coolant interactions
	Ex-vessel spreading

	Conclusions