metals

Article
Inspection of Prebaked Carbon Anodes Using
Multi-Spectral Acousto-Ultrasonic Excitation
Moez Ben Boubaker 1 , Donald Picard 2 , Carl Duchesne 1, *, Jayson Tessier 3 , Houshang Alamdari 4
and Mario Fafard 2
1 Aluminium Research Centre—REGAL, Département de Génie Chimique, Université Laval, Québec,
QC G1V 0A6, Canada; moez.ben-boubaker.1@ulaval.ca
2 Aluminium Research Centre—REGAL, Département de Génie Civil, Université Laval, Québec, QC G1V 0A6,
Canada; donald.picard@gci.ulaval.ca (D.P.); mario.fafard@gci.ulaval.ca (M.F.)
3 Alcoa Primary Metals Smelting Center of Excellence, Québec, QC G0A 1S0, Canada;
jayson.tessier@alcoa.com
4 Aluminium Research Centre—REGAL, Département de Génie des Mines, de la Métallurgie et des
Matériaux, Université Laval, Québec, QC G1V 0A6, Canada; houshang.alamdari@gmn.ulaval.ca
* Correspondence: carl.duchesne@gch.ulaval.ca; Tel.: +1-418-656-5184; Fax: +1-418-656-5993

Received: 15 June 2017; Accepted: 5 August 2017; Published: 8 August 2017

Abstract: Reduction cell operation in primary aluminum production is strongly influenced by the
properties of baked anodes. Producing consistent anode quality is more challenging nowadays
due to the increasing variability of raw materials. Taking timely corrective actions to attenuate the
impact of raw material fluctuations on anode quality is also difficult based on the core sampling
and characterization scheme currently used by most anode manufacturers, because it is applied on
a very small proportion of the anode production (about 1%), and long-time delays are required for
lab characterization. The objective of this work is to develop rapid and non-destructive methods
for the inspection of baked anodes. Previous work has established that sequential excitation of
smaller parts collected from an industrial sized anode using acousto-ultrasonic signals at different
frequencies allowed detecting and discriminating anode defects (pores and cracks). This was
validated qualitatively using X-ray computed tomography. This work improves the method
by using frequency-modulated excitation and building quantitative relationships between the
acousto-ultrasonic signals and defects extracted from tomography images using Wavelet Transforms
and Partial Least Squares (PLS) regression. The new excitation approach was found to provide similar
or better inspection performance compared with sequential excitation, while requiring a shorter
cycle time.

Keywords: baked carbon anode; non-destructive testing; acousto-ultrasonics; primary aluminum

smelting; PCA; PLS

1. Introduction
Most modern primary aluminum smelters use the Hall-Héroult (H-H) electrochemical process to
reduce alumina powder to metal aluminum. To carry out the reaction, prebaked carbon anodes are
utilized in order to distribute the electrical current in the reduction cells, and to supply the carbon
required by the reaction. Hence, the anodes are consumed and need to be replaced according to a
predefined set cycle. The overall performance of the H-H process is strongly influenced by the quality
of the anodes blocks. For instance, a low electrical resistivity is important for maximizing energy
efficiency. A high mechanical strength and density, and low reactivity to air and CO2 are desirable for
minimizing carbon consumption, and for reducing the environmental footprint and greenhouse gas
(GHG) emissions. Any physical defects within the anode blocks, such as pores, cracks or compositional

Metals 2017, 7, 305; doi:10.3390/met7080305 www.mdpi.com/journal/metals

Metals 2017, 7, 305 2 of 15

heterogeneities, may adversely affect the anode performance, and consequently, the performance of
the reduction cells. Producing high quality anodes consistently is therefore a major concern for the
manufacturers who are currently facing degrading quality, and increasing variability and cost of the
anode raw materials (petroleum coke and coal tar pitch).
The traditional anode quality control scheme used by the industry and consisting of collecting core
samples from the baked anodes and characterizing them in the laboratory is inadequate to address the
current needs. Indeed, core samples are gathered from a very small proportion of the anode production
(often less than 1%) because of the destructive and time/resource consuming nature of the procedures.
Furthermore, the samples themselves represent only 0.1–0.2% of the anode block volume, which has
properties known to be anisotropic. Finally, the core sample properties are typically available only after
a long time delay, which limits the implementation of feedback corrective adjustments to the anode
manufacturing process when deemed necessary. Therefore, alternative rapid and non-destructive
techniques are required to assess the quality of individual baked anodes before they are set in the H-H
reduction cells.
Recent research efforts focused on developing devices for measuring the electrical resistivity of the
individual green [1–3] and baked [4,5] anode blocks using different technologies. These sensors allow
measurement of a very important anode property related with energy efficiency, but their capacity
to detect, locate and diagnose physical defects within the anode blocks has yet to be established.
Alternatively, the performance of acousto-ultrasonic (AU) techniques for detection of various defects
in carbon anodes has been demonstrated in previous work [6] using sequential acoustic excitation
at different frequencies. Sliced anode samples were tested using this approach and their acoustic
responses was analyzed using Principal Component Analysis (PCA). The relationship between the
attenuation of the acoustic signals and the presence of different defects was validated qualitatively
using X-ray images collected from the slices. However, sequential excitation at multiple points on the
material is a lengthy process, and shortening the cycle time of the acoustic inspection scheme is highly
desirable for industrial implementation.
The aim of this work was to demonstrate that exciting the anode materials using a single frequency
modulated acoustic wave (multi-frequency signal) can lead to similar or better results compared to
sequential excitation, while significantly reducing cycle time. For instance, the sequential excitation
scheme used in previous work to obtain good inspection results consisted of testing the materials
at seven excitation frequencies in the 100–250 kHz range for each measurement point [6]. A single
frequency modulated wave in the same range would reduce cycle time by 7 to 1 for the same number
of data points. A side objective of this work was to validate the proposed approach more quantitatively,
by building empirical regression models between the acoustic response of the materials and the
corresponding Computed Tomography-scan images. These images were used for the sole purpose of
confirming the inspection results, as they would not be available in an industrial implementation.
Under frequency-modulated excitation, the material response signal is more complex as it also
contains multiple frequencies. To quantify the acoustic attenuation at different frequencies, the
signal needs to be first decomposed. One approach consists of using the Fast-Fourier transform
(FFT) [7–10]. However, this method only performs frequency domain decomposition and does not
capture variations in the signal frequency content through time, which may limit performance in defect
detection and identification. An alternative approach is to perform time-frequency decomposition of
the signals using Wavelet Transforms (WT) [11–14], which were shown to be a powerful tool in all
areas dealing with transient signals [15–17]. Qi et al. [18,19] showed the effectiveness of the Discrete
Wavelet Transform (DWT) for processing acousto-ultrasonic signals collected from composite materials.
A similar approach was used in this work.
To quantify defects in tomographic images, texture analysis techniques were used, since the
defects introduce local variations in grey level intensity in the images according to some relatively
well-defined patterns (e.g., round spots for pores and streak lines for cracks). Again here, one could
extract textural features from tomographic images using the two-dimensional Fast Fourier Transform
Metals 2017, 7, 305 3 of 15

(2D-FFT) [20]. The frequency decomposition of the images could be used to determine the severity
of the defects. However, variations in frequency content within the images (i.e., spatial information)
cannot be2017,
Metals extracted
7, 305 using this approach. On the other hand, the 2D Discrete Wavelet Transform 3 of 15

(2D-DWT) is effective for extracting both frequency and spatial information [21]. Wavelet texture
the defects. However, variations in frequency content within the images (i.e., spatial information)
analysis has proved a useful tool in several areas dealing with noise and low variation in images.
cannot be extracted using this approach. On the other hand, the 2D Discrete Wavelet Transform (2D-
After applying the WT to both sets of data (acoustic signals and images), wavelet features were
DWT) is effective for extracting both frequency and spatial information [21]. Wavelet texture analysis
extracted and
has proved analyzed
a useful using
tool inPCA in order
several to explore
areas dealing withthe clustering
noise and lowpattern ofin
variation the various anodes slices.
images.
A regression
Aftermodel was
applying thethen
WT built
to bothbetween both
sets of data data sets.
(acoustic Theand
signals results showed
images), that
wavelet the attenuated
features were
acoustic response
extracted and of the materials
analyzed obtained
using PCA after
in order frequency-modulated
to explore excitation
the clustering pattern of thewas sensitive
various anodesto the
slices.
presence of A regression
cracks withinmodel
anodewas then built
samples, and tobetween both data
the density sets.distributed
of pores The resultsthroughout
showed thatthe theanode
attenuated acoustic response of the materials obtained after frequency-modulated
samples (slices). The performance of this approach was found to be similar to that obtained using excitation was
sensitive
sequential to the presence
excitation [6], but of cracks
with within anode
a significant samples,
reduction in and
cycletotime.
the density of pores distributed
throughout the anode samples (slices). The performance of this approach was found to be similar to
that obtained
2. Materials using sequentialData
and Experimental excitation [6], but with a significant reduction in cycle time.
Acquisition

2. Materials
2.1. Baked Anode and Experimental
Samples and X-rayData Acquisition
Images
A
2.1.full-size industrial
Baked Anode Samples baked
and X-ray carbon
Imagesanode was scanned using a Siemens Somatom Sensation
64 tomograph (512 × 512 pixels) (Siemens Healthcare GmbH, Erlangen, Germany). The scanning area
A full-size industrial baked carbon anode was scanned using a Siemens Somatom Sensation 64
of thetomograph
X-ray apparatus was limited to objects of lessGmbH, 40 × 40 mm
than a Erlangen, 2 cross-section. Since the anode
(512 × 512 pixels) (Siemens Healthcare Germany). The scanning area of
was too
the large
X-ray to fit in thewas
apparatus scanning
limited area in one
to objects of piece, it was
less than a 40 cut
× 40inmmseveral slices, asSince
2 cross-section. shownthe in Figure 1.
anode
The cutting
was toowas largeperformed
to fit in the with
scanning carearea
to avoid
in onealtering any cut
piece, it was anode defects.
in several Furthermore,
slices, as shown in for a single
Figure
CT image
1. Thereconstruction,
cutting was performed a higher with image
care toresolution is obtained
avoid altering any anode when theFurthermore,
defects. samples are forsmaller in size
a single
CT image reconstruction, a higher 2
image resolution is obtained when the
(down to a cross-section of 50 × 50 mm where the resolution limit of the tomograph used is reached), samples are smaller in size
which(down to a cross-section ofbetween
50 × 50 mm 2 where the resolution limit of the tomograph used is reached),
increases the contrast defects and the background [22]. The anode was first cut into
which increases the contrast between defects and the background [22]. The anode was first cut into
26 slices (Figure 1a), and then each of them were cut further in halves along their heights (Figure 1b).
26 slices (Figure 1a), and then each of them were cut further in halves along their heights (Figure 1b).
The top and bottom of each half slice was also cut to obtain samples of uniform geometry before
The top and bottom of each half slice was also cut to obtain samples of uniform geometry before
collecting the acoustic response of the material. Note that the following eight slices were selected for
collecting the acoustic response of the material. Note that the following eight slices were selected for
further AU analysis:
further AU analysis:2, 3,2,5,3,7,5,11,7, 15,
11, 24
15,and 25 (indicated
24 and 25 (indicatedusing redred
using numbers
numbers ininFigure
Figure1a).
1a).These
Theseslices
spanned a range of internal structures, and were selected from areas which
slices spanned a range of internal structures, and were selected from areas which were expected were expected to contain
to
different types of defects, such as pores and cracks (see [6] for more
contain different types of defects, such as pores and cracks (see [6] for more details). details).

FigureFigure
1. The1. The sliced
sliced baked
baked anode.(a)
anode. (a)the
theselected
selected 88 slices
slicesare
areidentified in in
identified red; (b) (b)
red; example of a slice
example of a slice
used for acousto-ultrasonic testing; (c) example X-ray image obtained for slice
used for acousto-ultrasonic testing; (c) example X-ray image obtained for slice 7. The corridors 7. The corridors are are
identified by the red lines and numbered from 1–6. The black region corresponds
identified by the red lines and numbered from 1–6. The black region corresponds to one of the anode to one of the anode
slots. Adapted from [6].
slots. Adapted from [6].
Metals 2017, 7, 305 4 of 15
Metals 2017, 7, 305 4 of 15

CT-scan measured
The CT-scan measuredthe theattenuation
attenuationcoefficients
coefficients
of of
thethe material
material for for
a seta of
set512 512×3D
of ×512 512 3D
voxel
voxel (volumetric pixel) per scan, which corresponded to the tomograph resolution.
(volumetric pixel) per scan, which corresponded to the tomograph resolution. Given the Given the slice
dimensions, the voxels hadhad aa resolution
resolutionof 0.7×× 0.6
of0.7 0.6 × 0.7 mm33.. The
× 0.7 The X-ray
X-ray attenuation
attenuation coefficients of
the voxels (also called Computed Tomography or CT numbers) were converted to grayscale images
showing spatial
spatial variations
variationscorrelated
correlatedwith
withthethematerials
materialsdensity
density[22,23].
[22,23].Cracks
Cracks and
and pores
pores appear
appear as
as groups
groups of of darkpixels
dark pixelssurrounded
surroundedby bylighter
lighter background
background (carbon
(carbon material). The The high-density
impurities appear as bright pixels inside a darker background. Figure Figure 1c shows an example X-ray
image for one slice.

2.2. Acousto-Ultrasonic
2.2. Acousto-Ultrasonic (AU)
(AU) Signal
Signal Acquisition
Acquisition
The acousto-ultrasonic
The acousto-ultrasonic inspection
inspection system
systemused usedininthis thiswork
workwas wasdescribed
described in in detail
detail in in
[6].[6].
It
It consisted
consisted ofof a multi-channelacousto-ultrasonic
a multi-channel acousto-ultrasonicapparatusapparatususing usingtwo twoacoustic
acousticsensors
sensors (an
(an emitter
emitter and
and
aa receiver),
receiver), aa pre-amplifier,
pre-amplifier, and and aa software
software for for controlling
controlling the the excitation
excitation andand recording
recording the
the emission
emission
signals. After testing several sets of sensors, it was found that using transducers having a
signals. After testing several sets of sensors, it was found that using transducers having a resonant
resonant
frequency at 55 kHz provided the best results for the tested materials.
frequency at 55 kHz provided the best results for the tested materials. The sensors were mounted on The sensors were mounted
on both
both ends
ends of of
thethe anode
anode slices
slices (one
(one onontop
topandandthe theother
otheron onthethe bottom
bottom surface)
surface) using
using mechanical
mechanical
clamps. Coupling
clamps. Coupling gel gel was
was applied
applied on on the
the anode
anode surfaces
surfaces before
before the the sensors
sensors were
were putput into
into contact
contact with
with
the materials. This was made to ensure good contact between the
the materials. This was made to ensure good contact between the sensors and the anode samples, sensors and the anode samples,
which
which typically
typicallyhavehaveroughroughsurfaces.
surfaces.TheThesensors
sensorswerewereplaced vis-à-vis
placed each other,
vis-à-vis eachand moved
other, andtogether
moved
along thealong
together anode thewidth
anodetowidth
measure the AU the
to measure response of the materials
AU response included
of the materials in the in
included sixthe
so-called
six so-
corridors (identified by white chalk in Figure 1b, and by red lines in Figure
called corridors (identified by white chalk in Figure 1b, and by red lines in Figure 1c). The material 1c). The material structure
clearly changed
structure clearly from
changedthe center
from the of the anodes
center of the to anodes
the external
to thesurface
external (corridor
surface 1(corridor
and 6 in1Figure
and 6 1c,in
respectively), as shown by variations in material density (i.e., gray level)
Figure 1c, respectively), as shown by variations in material density (i.e., gray level) and the presence and the presence of pores
and
of cracks.
pores andAs mentioned
cracks. in the previous
As mentioned section, such
in the previous variations
section, in the material
such variations structurestructure
in the material are also
expected
are along thealong
also expected length theoflength
the anode
of the(i.e.,
anodevarious
(i.e., slices).
various slices).
The excitation signal was specified as a modulated frequency
The excitation signal was specified as a modulated frequency signalsignal between
between 100 100 and
and 300
300 kHz
kHz
as shown in Figure 2. This range was selected based on previous
as shown in Figure 2. This range was selected based on previous work [6]. The AU measurements work [6]. The AU measurements
were
were mademade using
using aa pre-amplification
pre-amplification gain gain of of 4040 dB,
dB, and
and sampling
sampling rate rate of
of 1000
1000 kHz.
kHz. The
The selected
selected
sampling rate is sufficiently fast to avoid aliasing at frequencies lower than
sampling rate is sufficiently fast to avoid aliasing at frequencies lower than 300 kHz. This procedure 300 kHz. This procedure
was
was repeated
repeated forfor all
all six
six corridors
corridors (Figure
(Figure 1b,c)
1b,c) inin each
each of of the
the selected
selected anode
anode slices. Note that
slices. Note that the
the type
type
of transducer
of transducer andand measurement
measurement settings
settings maymay needneed to to be
be slightly
slightly adjusted
adjusted ifif the
the system
system isis to
to be
be used
used
for inspecting anodes from different
for inspecting anodes from different manufacturers.manufacturers.

Figure
Figure 2.
2. The
The frequency
frequency modulated
modulated waveform
waveform used
used as
as the
the excitation
excitation signal.
signal. Both
Both the
the time
time series
series (top)
(top)
and the frequency content (bottom) of the signal are shown.
and the frequency content (bottom) of the signal are shown.
Metals 2017, 7, 305 5 of 15

3. Extraction of Features from Acoustic Signals and X-ray Images

3.1. Acousto-Ultrasonic Signals

The majority of the acousto-ultrasonic signals propagating through the porous anode materials
are in the frequency range of 100–300 kHz. Dividing this range into a set of consecutive frequency
bands may help relate different time-frequency features of AU signals to various anodes defects.
The wavelet transforms were used to decompose the signals and extract meaningful time-frequency
domain features from them.
The decomposition of a continuous 1D signal x(t) using a specific scaled or dilated mother wavelet
Ψ (t,a,b) is obtained by the convolution of both the signal and the wavelet function as follows [21]:

+∞
t−b
 
1
Z
d a,b = √ x (t) Ψ dt (1)
a a
−∞

In the above Equation, Ψ (t) is the mother wavelet function, which corresponds to a wave
of finite length and having a particular shape. Several types of mother wavelets exist (e.g., Haar,
Daubechies, Coiflet, Symlet, Mexican Hat, etc.), and they mainly differ by their shape (frequency
content). The type of mother wavelet is typically selected to match the shape of the signal x(t) the best
possible. The Daubechies wavelet (Db5) was selected in this work. This choice came after selecting the
best results from trying several different mother wavelets. Parameters a and b are called the scaling
and the translation parameters, respectively. The former stretches the mother wavelet and changes
its frequency content, while the latter performs a translation of the wavelet over time. The result
of the convolution is a scalar quantity da ,b , called the wavelet detail coefficient. It represents how
well the signal x(t) matches the dilated mother wavelet at scale a and time point b. Changing the
scaling coefficient a from a small to a large value progressively dilates the wavelet and therefore
allows extracting information about signal x(t) from high to low frequency. Translating the wavelet
using parameter b from the beginning to the end of the x(t) signal allows information to be captured
at different time points. Hence, changing a and b in a nested fashion performs a spatial-frequency
decomposition of signal x(t), and how well the signal matches the wavelet at scale a and time point b is
quantified by the detail coefficient da ,b . Note that the acoustic signals analyzed in this work are discrete
(i.e., sampled at a given frequency). A discrete version of Equation (1) exists for discrete signals as
well. This leads to the so-called Discrete Wavelet Transform (DWT). Values taken by parameters a and
b are constrained for computing efficiency and to avoid signal aliasing. For more details on the DWT,
the reader is referred to [21].
After performing the wavelet decomposition of the discrete signal x(i), a set of vectors of detail
coefficients da (i) were obtained, one for each scale a. Each vector contained the detail coefficient at a
given scale but for all time points. In other words, the original signal x(i) was decomposed into a set
of detail signals da (i), each containing information in a certain frequency band. Three time domain
features were then calculated from each time series da (i), i = 1, 2, . . . , N, N being the number of samples
in the time series. These features are the maximum (MAX), the root mean square (RMS) and the time
of flight, or arrival time (AT). These features are calculated as follows:

MAXa = maxiN=1 d a (i ) (2)

r
1 N
N ∑ i =1
RMSa = [d a (i )]2 (3)

ATa = i when d a (i ) ≥ 0.15 × max(d a (i )) (4)

The acoustic response of the materials x(i) was decomposed up to level 4 (four different discrete
values of a) and the MAX and RMS features were calculated from each of them as well as the arrival
Metals 2017, 7, 305 6 of 15

time. Figure 3 provides an example of a raw acoustic signal decomposed into 4 frequency bands
(details
Metals 1–4),
2017, 7,and
305 also shows the residual signal after four decomposition levels (called approximation).
6 of 15

Figure 3. Example wavelet decomposition of an acousto-ultrasonic (AU) signal using 4 decomposition

Figure 3. Example wavelet decomposition of an acousto-ultrasonic (AU) signal using 4 decomposition
levels. The raw signal and the four wavelet detail signals as well as the residuals (approximation)
levels. The raw signal and the four wavelet detail signals as well as the residuals (approximation)
are presented.
are presented.
The data were then collected in a (48 × 15) dimensional matrix X. The 15 columns correspond to
theThe data
three were then
features collected
computed frominthe four×wavelet
a (48 15) dimensional
detail signals, matrix X. The
as well 15 columns
as the approximationcorrespond
for all to
theofthree features computed from the four wavelet detail signals,
the 48 samples (six corridors × eight anode slices). Signal and image processing were performed as well as the approximation for all
of using
the 48Matlab
samples (six corridors
version × eight anode
R2014a (MathWorks, slices).
Natick, MA,SignalUSA).and image processing were performed
using Matlab version R2014a (MathWorks, Natick, MA, USA).
3.2. X-ray Image Texture Analysis
3.2. X-ray Image Texture Analysis
The texture of X-ray images was analyzed using a very similar approach as for the acoustic
signals. The tomography
The texture of X-ray images datawas were first converted
analyzed using a very to grayscale
similar approachimages. as The
for Discrete
the acoustic Wavelet
signals.
TheTransform
tomography was datathen were
applied firsttoconverted
them. Theto main difference
grayscale images. in the Theapplication of wavelets
Discrete Wavelet to images
Transform was
and acoustic signals is that images were 2D signals. An image
then applied to them. The main difference in the application of wavelets to images and acoustic is a matrix of data I(n X ,n Y ) where n X

and n are the number of pixels in the horizontal and vertical

signals is that images were 2D signals. An image is a matrix of data I(nX ,nY ) where nX and nY are
Y directions in the images and the
theelements
number of that matrix
of pixels in thecorrespond
horizontal and to the X-ray directions
vertical attenuationincoefficients
the images converted to grayscale
and the elements of that
values. The convolution between a mother wavelet and the image
matrix correspond to the X-ray attenuation coefficients converted to grayscale values. The convolution signals (i.e., discretized version of
Equation (1)) is performed along each row and each column of the
between a mother wavelet and the image signals (i.e., discretized version of Equation (1)) is performed image (horizontally and vertically)
as well as diagonally using a filter bank [21]. Wavelet detail coefficients were obtained for each pixel
along each row and each column of the image (horizontally and vertically) as well as diagonally using
of the images at each scale and for each of the three directions. These were collected into matrices Djk
a filter bank [21]. Wavelet detail coefficients were obtained for each pixel of the images at each scale
where j corresponds to the scale, k to each direction (h,v,d), and each element of these matrices match
and for each of the three directions. These were collected into matrices Dj k where j corresponds to the
the detail coefficient at a certain pixel location (nX, nY). These matrices can also be displayed as images
scale, k to each direction (h, v, d), and each element of these matrices match the detail coefficient at a
for visual interpretation, and are typically called wavelet detail sub-images (Djh, Djv, Djd). At the first
certain pixel location (n X , n Y ). These matrices can also be
scale, high frequency information is captured from the images in all three directions, displayed as images for visualand interpretation,
these are
and are typically called wavelet detail sub-images (D h , D v , D d ). At the first scale, high frequency
associated with small objects or patterns in the image. j j
As j scale increases, lower frequency
the
information
information is introduced
captured from by largerthe images
objects in or all three is
patterns directions,
extracted.and The these are associated
5th order Daubechieswith small
wavelet
objects or patterns in the image. As the scale increases,
(Db5) wavelet was selected for extracting textural information from the X-ray images. lower frequency information introduced by
larger objects or patterns is extracted. The 5th order Daubechies wavelet
Several scalar textural descriptors or textural features can be calculated from the wavelet detail (Db5) wavelet was selected
forsub-images,
extracting textural
the mostinformation
popular being fromthethe X-rayvariance,
energy, images.entropy, and the mean [24,25]. Our results
Several
showed scalar
that the textural
energy isdescriptors
sufficient toorextract
textural features can
information be calculated
related with anode from the wavelet
defects (pores anddetail
sub-images,
cracks) withinthe most popular
the X-ray images. being
Thetheenergy
energy,isvariance,
calculatedentropy, as follows andfrom
the mean [24,25].detail
the wavelet Our results
sub-
images
showed at each
that scale and
the energy for the three
is sufficient directions:
to extract information related with anode defects (pores and cracks)
within the X-ray images. The energy is calculated nX n Y as follows from the wavelet detail sub-images at

2
each scale and for the three directions: D k
j ( m , n )
E jk = mn=1 n =1n  (5)
nY XnYk
X
2
∑ ∑ D j (m, n)

k m =1 n =1
where Dj is the detail coefficient for E j = located at the (nX, nY) position within an image in the kth(5)
pixel
k
n X nY
direction (k = h, v, d) and at the jth scale. The total number of pixels within the images (nX × nY) was
used to normalize the energy values. The X-ray images were decomposed up to scale 4. This yielded
Metals 2017, 7, 305 7 of 15

where Dj k is the detail coefficient for pixel located at the (nX , nY ) position within an image in the kth
direction (k = h, v, d) and at the jth scale. The total number of pixels within the images (nX × nY ) was
used Metals
to normalize
2017, 7, 305 the energy values. The X-ray images were decomposed up to scale 4. This 7 ofyielded
15
12 energy values per anode corridor (four scales × three directions). These values were collected for
the 4812corridors
energy values per anode corridor (four scales × three directions). These values were collected for
and stored in matrix Y of dimensions (48 × 12).
the 48 corridors and stored in matrix Y of dimensions (48 × 12).
The application of wavelet decomposition to one such X-ray image is shown in Figure 4.
The application of wavelet decomposition to one such X-ray image is shown in Figure 4. It
It displays thethe
displays original image
original imageas aswell
well as
as the detailsub-images
the detail sub-images forfor each
each of the
of the fourfour scales
scales and directions.
and directions.
The sub-images
The sub-images areare
identified
identifiedby bythe
the direction first(h,(h,
direction first v, and
v, d) d) and
thenthen
by theby thenumber
scale scale number
(1,2,3,4). (1,2,3,4).
For
For instance,
instance, sub-images
sub-images v3 v3 display
displaythe theinformation
information extracted
extracted fromfrom
the the image
image by wavelet
by the the wavelet
in thein the
vertical direction
vertical directionat scale 3. The
at scale 3. Thevertical
verticaland
anddiagonal sub-imagesatatscales
diagonal sub-images scales
2–42–4 clearly
clearly capture
capture the the
profile line of the vertical and diagonal cracks in the image. This information
profile line of the vertical and diagonal cracks in the image. This information was regressed against was regressed against
the acoustic
the acoustic signalssignals in order
in order to link
to link thethe acousticattenuation
acoustic attenuation to toanode
anodedefects
defectsquantitatively.
quantitatively.

Figure 4. 2D4.Discrete
Figure wavelet
2D Discrete decomposition
wavelet decompositionof of an X-rayimage
an X-ray imageatatfour
four scales
scales andand in three
in three directions.
directions.
The wavelet
The wavelet sub-images
sub-images areare identifiedby
identified bythe
the direction
direction of
of analysis
analysis(h,(h,v, v,
d) d)
followed
followedby the
by scale
the scale
number
number (1–4).(1–4).

4. Multivariate Statistical Methods for Analysis of the Acousto-Ultrasonic Signal and Image
4. Multivariate Statistical Methods for Analysis of the Acousto-Ultrasonic Signal and
ImageFeatures
Features
Multivariate statistical methods were used to analyze the large data matrices obtained in this
Multivariate statistical methods were used to analyze the large data matrices obtained in this
study. First, PCA was used to cluster the X-ray images based on their texture (Y data). Then, Partial
study.Least
First, PCA was used to cluster the X-ray images based on their texture (Y data). Then, Partial
Squares (PLS) regression was utilized to build relationships between the acoustic attenuation
Leastdata
Squares (PLS)
(X) and thenregression wasfeatures
image textural utilized toThese
(Y). buildmethods
relationships between
are briefly the below.
described acoustic attenuation
data (X) and then image textural features (Y). These methods are briefly described below.
4.1. Principal Component Analysis
4.1. Principal Component Analysis
Data clustering in the latent variable space using methods such as PCA is widely used for the
Data clustering
analysis in the latent
of large datasets variable
containing noisyspace usingcollinear
and highly methods such
data. PCAassummarizes
PCA is widely used for the
the variations
contained
analysis in datasets
of large a large setcontaining
of J variables X = [xand
noisy 1, x2,highly
…, xJ] by a much data.
collinear smaller number
PCA of orthogonal
summarizes latent
the variations
variables (or scores) T = [t 1, t2, …, tA] (A << J) together capturing the dominant sources of variations in
contained in a large set of J variables X = [x1 , x2 , . . . , xJ ] by a much smaller number of orthogonal
latentthe data. The(or
variables scores are obtained
scores) T = [t1 ,by t2an
, . .eigenvector-eigenvalue
. , tA ] (A << J) together bilinear decomposition
capturing of the X matrix
the dominant sources of
and is mathematically expressed as follows [26,27]:
variations in the data. The scores are obtained by an eigenvector-eigenvalue bilinear decomposition of
A
the X matrix and is mathematically expressed as follows [26,27]:
X =  t a p aT + E = TP T + E (6)
a =1
A
where the orthogonal vectors ta are X=the∑ T
ta pvariables
latent TPT +scores)
a + E = (called E providing the coordinates of (6)
each sample in the low dimensional subspacea=1 (plane or hyperplane) after projection. The subspace
itself is defined by the A orthonormal loading vectors pa, which are linear combinations of the original
where the orthogonal vectors ta are the latent variables (called scores) providing the coordinates of
variables (i.e., ta = X pa). The projection residuals are collected in the residual matrix E. The loading
each sample in the low dimensional subspace (plane or hyperplane) after projection. The subspace
vectors are calculated in such a way that t1 explains the greatest amount of variance in X, t2 the second
itself is defined by the A orthonormal loading vectors pa , which are linear combinations of the original
greatest amount of variance not explained by the first component, and so on. Plots of the score values
variables (i.e., ta = X pa ). The projection residuals are collected in the residual matrix E. The loading
Metals 2017, 7, 305 8 of 15

vectors are calculated in such a way that t1 explains the greatest amount of variance in X, t2 the second
greatest amount of variance not explained by the first component, and so on. Plots of the score values
are used to visualize the clustering pattern of the data while the loadings allow interpretation of the
patterns by the changes in the variables. The reader is referred to [27] for more details on PCA.

4.2. Partial Least Squares Regression

The PLS regression method was used here to correlate the acousto-ultrasonic features of the anode
corridors gathered in the X matrix with corresponding image textural features collected in Y. Similar
as for PCA, this regression method has an advantage of overcoming the strong correlation between
features in both data blocks. Each PLS latent variable is calculated in such a way as to maximize the
covariance in both data matrices and the following model structure is obtained [28,29]:

X = TPT + E (7)

Y = TCT + F (8)
−1
T = XW* = XW(PT W) (9)

where the latent variable space of both X and Y corresponds to the scores T (orthogonal PLS
components) which relate X to Y. The sub-spaces of X and Y are modeled by the loading matrices
P and C, respectively, while the corresponding residuals are stored in E and F. Finally, the so-called
weight vectors W* contain the linear combinations that express T in terms of X and predictions of Y to
be made.
The PLS method also performs clustering of samples vector in the latent variable subspace
according to the Y data. The clustering patterns can be visualized using scatter plots of the scores
(t’s). The differences between the clusters can be interpreted using the loading weight vectors (w*’s).
The reader is referred to Wold et al. [29] for more details about these methods. The ProMV software
version 15.08 (ProSensus, Ancaster, ON, Canada) was used to build PCA and PLS models.

5. Results and Discussion

The X-ray image textural features (Y data) were analyzed first using PCA in order to associate
them with anode defects (voids such as pores and cracks). A PLS regression model was then built using
both the acoustic and image textural features in order to correlate them. This provided a quantitative
validation of the impact of anode defects on the acousto-ultrasonic features.

5.1. Texture Analysis of X-ray Images

Using the cross-validation procedure [30], two PCA components (or latent variables) were
found to be significant, and together explained most of the variations in the image textural features.
The cumulative sum of squares explained (R2 ) and predicted (Q2 ) by the first two PCA components
are provided in Table 1. The first two components were found to be sufficient to discriminate the
corridors based on their textural features since they explained over 83% of variance in the textural
features stored in Y (Q2 = 83.21% in Table 1).

Table 1. Percent cumulative sum of squares explained (R2 ) and predicted (Q2 ) by the Principal
Component Analysis (PCA) model built on textural features (Y data) collected from X-ray images of
anode slices.

Component R2 (%) Q2 (%)

1 60.3 58.7
2 84.0 83.2
Metals 2017, 7, 305 9 of 15

The main variability directions in the image dataset captured by the two PCA components are
presented in the t1 –t2 score plot shown in Figure 5. Each marker in this plot represents a summary of
the textural features of one particular corridor. These are identified by the slice number followed by
the corridor number.
Metals 2017, 7, 305 Corridors having similar textural features cluster close to each other, while 9 of 15 those
having different features cluster in different locations in the latent variable space. It was observed
those
that the having different
corridors located infeatures cluster
the same in different
position within locations in cluster
the slices the latent variable(see
together space. It wasof the
markers
observed that the corridors located in the same position within the slices cluster
same color in Figure 5). Furthermore, each group was discriminated from the others along both PCA together (see markers
of the same color in Figure 5). Furthermore, each group was discriminated from the others along both
components, except those of slice number 25. This slice was located close to the corner of the anode
PCA components, except those of slice number 25. This slice was located close to the corner of the
block, and contained larger pores and cracks, which made them similar to either corridors #1–2, which
anode block, and contained larger pores and cracks, which made them similar to either corridors
typically contained several cracks, or to corridors #6 that contained many pores. This explained why
#1–2, which typically contained several cracks, or to corridors #6 that contained many pores. This
the corridors
explainedof why slice
the#25 clustered
corridors close
of slice #25toclustered
corridors #1–2
close or #6. #1–2 or #6.
to corridors

Figure 5. Score plot of the PCA model (t1–t2) showing the clustering pattern of the corridors in each
Figure 5. Score plot of the PCA model (t1 –t2 ) showing
slice based on their textural features extracted
the clustering pattern of the corridors in each
from X-ray images. Observations are identified by the
slice slice
based on their
number (redtextural
numbers features
in Figure extracted
1) followedfrombyX-ray images.
the corridor Observations
number are identified
(1–6). Different colors areby the
slice used
number (red numbers
to distinguish in Figure
the corridors 1) followed
as shown by the
in the anode corridor
slice numberin
image included (1–6). Different colors are
the figure.
used to distinguish the corridors as shown in the anode slice image included in the figure.
The clustering pattern in the score space can be interpreted based on changes in the textural
features by examining the p1–p2 loading plot shown in Figure 6. Each point in the loading plot
The clustering pattern in the score space can be interpreted based on changes in the textural
corresponds to one textural feature calculated from the image of each corridor. They are labeled
features by examining the p1 –p2 loading plot shown in Figure 6. Each point in the loading plot
according to the feature name followed by the wavelet decomposition level. The loading values
corresponds
define thetoweightone textural featurefeatures
of each textural calculated in eachfrom the image
component. Theofhigher
each the
corridor.
absoluteThey
valueare labeled
of the
according
loadings the more important the features are in a given component. The signs of the loadingsdefine
to the feature name followed by the wavelet decomposition level. The loading values
the weight
indicates of the
each
signtextural
of theirfeatures in each
correlation. component.
Features The higher
having loading valuesthe absolute
of the same sign value
are of the loadings
positively
the more important
correlated, and thosethe having
features are in signs
opposite a given are component. The signs
negatively correlated. Theofloading
the loadings indicates
plot reveals that the
the first component is driven by the features in most wavelet decomposition
sign of their correlation. Features having loading values of the same sign are positively correlated, levels, and all of them
and have
thoseloading
havingvaluesopposite of the
signssame
aresign (i.e., theycorrelated.
negatively vary all together in the plot
The loading samereveals
direction).
thatThis
the first
component seems to explain variations in density and/or the
component is driven by the features in most wavelet decomposition levels, and all of them total volume of voids (i.e., defect have
severity) across the X-ray images. A denser region is characterized by higher CT numbers, and hence
loading values of the same sign (i.e., they vary all together in the same direction). This component
by higher grey levels in the image. This typically leads to higher energy values as calculated from the
seems to explain variations in density and/or the total volume of voids (i.e., defect severity) across
wavelet texture analysis. Since corridors 3–4 (and 5 to some extent) were known to be denser, it was
the X-ray
expectedimages. A denser
that these corridorsregion
wouldis characterized by higher
fall in the positive t1 regionCT(see
numbers,
Figure 5)and hence
because thebyenergies
higher grey
levelsatin
allthe image.
scales This typically
and directions wouldleads to higher
be higher. On theenergy values
contrary, as calculated
corridors 1–2 and 6from the wavelet
typically contained texture
more defects (cracks and pores), and therefore are were less dense, with lower CT numbers, grey
 Metals 2017, 7, 305 10 of 15

level intensities and energies. Thus, it was expected that these would fall in the negative t1 region of
the score plot. This clustering pattern along t1 is indeed observed in Figure 5.
Metals 2017, 7, 305 10 of 15
The second component, however, was driven by a contrast between the energies at scales 1–2
vs. 3–4, since these two groups have loading values of opposite signs. This component seemed to
capture textural variation that corresponded to defects size. Corridors containing larger defects (e.g.,
analysis. Since corridors 3–4 (and 5 to some extent) were known to be denser, it was expected that
cracks) should be characterized by higher energies at lower frequencies (scales 3–4) and lower energy
these corridors would fall in the positive t1 region (see Figure 5) because the energies at all scales and
at higher frequencies (scales 1–2) since the lower frequency bands are associated with larger objects
directions would
or patterns. be higher.
Since corridorsOn 1–2the contrary,
typically corridors
contained more1–2 and they
cracks, 6 typically contained
fell in the positive more defects
t2 region
(cracks and pores), and therefore are were less dense, with lower CT numbers, grey level
(higher energies at scales 3–4 and lower at scales 1–2). The situation was the opposite with corridor 6 intensities
and energies. Thus, it was
typically containing poresexpected
(clusters inthat these twould
negative fall
2 region). in the
Hence, thenegative
second PCAt1 region of the
component score plot.
seemed
This clustering pattern
to discriminate alongoft1defects
the types is indeed
in theobserved in Figure
anode samples (i.e., 5.
pores vs. cracks).

Figure 6. Loading
6. Loading plotplot of the
of the PCA
PCA model(p(p1–p
model –p2) built on image textural features (Y data). The energy
Figure 1 2 ) built on image textural features (Y data). The energy
features are identified by letter “E” followed
features are identified by letter “E” followed by by the
the wavelet
waveletdirection (H,
direction V, V,
(H, or or
D),D),
andand
scalescale
number.
number.

To conclude this section, the textural features extracted from the X-ray images appeared to
The second
capture component,
relevant however,
information wastype
about the driven by defects
of the a contrast between
present in thethe energies
anode slicesatand
scales 1–2 vs.
their
3–4, since these two groups have loading values of opposite signs. This component
severity. Thus, they can be used to perform supervised clustering of the AU signals collected from seemed to capture
textural
the variation
corridors, that corresponded
and allow to defects
for a quantitative size. Corridors
assessment containingoflarger
of the sensitivity defects
AU signal (e.g., to
features cracks)
anode
should defects.
be characterized by higher energies at lower frequencies (scales 3–4) and lower energy at higher
frequencies (scales 1–2) since the lower frequency bands are associated with larger objects or patterns.
Since5.2. Regression of X-ray Image Textural Features on Acousto-Ultrasonic Features
corridors 1–2 typically contained more cracks, they fell in the positive t2 region (higher energies at
scales 3–4Five
andPLSlowercomponents were The
at scales 1–2). found to be statistically
situation significant
was the opposite by acorridor
with standard cross-validation
6 typically containing
poresprocedure
(clusters [30], and together
in negative explained
t2 region). aboutthe
Hence, 40% of the variance
second of the image
PCA component features
seemed to(Y data). The the
discriminate
typesgoal here was
of defects in to use
the the Ysamples
anode data to perform a kind
(i.e., pores vs.ofcracks).
supervised clustering of the AU signals based
on defects found in the X-ray images, and not to predict the image features (i.e., energies) with high
To conclude this section, the textural features extracted from the X-ray images appeared to capture
accuracy. The first two PLS components were found to be sufficient to discriminate the samples based
relevant information about the type of the defects present in the anode slices and their severity. Thus,
on their acousto-ultrasonic features, and therefore, only these components will be discussed in
they can be used to perform supervised clustering of the AU signals collected from the corridors, and
this section.
allow for The
a quantitative
latent variable assessment of for
score space the the
sensitivity
first two of AU signaloffeatures
components the PLSto anode
model is defects.
displayed in
Figure 7. Each marker in the plot corresponds to the acousto-ultrasonic response of one corridor,
5.2. Regression of X-ray Image
which are identified Textural
as described Features
in the on Acousto-Ultrasonic
previous Features
section. The clustering pattern in the t1–t2 score space
reveals
Five PLSthat corridors 1–2,
components 3–5 found
were and 6 of
to any slice cluster significant
be statistically in three groups
by a(black, orange
standard and blue
cross-validation
procedure [30], and together explained about 40% of the variance of the image features (Y data).
The goal here was to use the Y data to perform a kind of supervised clustering of the AU signals
based on defects found in the X-ray images, and not to predict the image features (i.e., energies) with
high accuracy. The first two PLS components were found to be sufficient to discriminate the samples
based on their acousto-ultrasonic features, and therefore, only these components will be discussed in
this section.
The latent variable score space for the first two components of the PLS model is displayed in
Figure 7. Each marker in the plot corresponds to the acousto-ultrasonic response of one corridor,
Metals 2017, 7, 305 11 of 15

which are identified as described in the previous section. The clustering pattern in the t1 –t2 score
space reveals that corridors 1–2, 3–5 and 6 of any slice cluster in three groups (black, orange and blue
markers, respectively). At this point, it is important to note that corridor #1 was located at the center of
the anode, and corridor #6 at the outer surface.
Metals 2017, 7, 305 11 of 15
The first component (t1 ) in Figure 7 seemed to capture the types of defects in the materials,
since themarkers,
corridors where most
respectively). of point,
At this the cracks were found
it is important to note(1–2) were located
that corridor in the negative
#1 was located at the center t1 region.
Those characterized
of the anode, and bycorridor
a high concentration of pores (corridor 6) are located in the positive t1 region,
#6 at the outer surface.
and corridors The3–5
firstcontaining
component (ta1)mixture
in Figure of 7 seemed
both fallto capture the types
in between theoftwo.
defects in the
The * -c –w * -c
w1materials, since
1 2 2 loadings
the corridors where most of the cracks were found (1–2) were located in the
bi-plot presented in Figure 8 confirms this interpretation. The signs of X-loadings and Y-loadings (w* ’s negative t 1 region. Those

characterized by a high concentration of pores (corridor 6) are located in the positive t1 region, and
and c’s or black and red markers in Figure 8) indicate the sign of the *correlation between pairs of
corridors 3–5 containing a mixture of both fall in between the two. The w1 -c1–w2*-c2 loadings bi-plot
featurespresented
whereas in their absolute
Figure values
8 confirms this is proportional
interpretation. Thetosigns
theirofimportance
X-loadings and in aY-loadings
given component.
(w*’s and Note
that comparing
c’s or blackthe andblack or the in
red markers red loading
Figure values
8) indicate theallows interpretation
sign of the of thepairs
correlation between correlation
of features between
the acoustic
whereas or their
the image
absolutetextural
values isfeatures,
proportional whereas
to theircomparing
importance inthe blackcomponent.
a given against the Note red loadings
that
providescomparing
information the black
aboutor the
the red loading values allows
cross-correlation between interpretation
the acoustic of theandcorrelation between
the textural the
features, and
acoustic or the image textural features, whereas comparing the black against the red loadings
the relationship between both the X and Y datasets. Clearly, the first PLS component captured a similar
provides information about the cross-correlation between the acoustic and the textural features, and
contrastthe
in relationship
the energiesbetween at scales 3–4 vs. 1–2 as the second PCA component discussed in the previous
both the X and Y datasets. Clearly, the first PLS component captured a
section.similar
The variations in the acoustic
contrast in the energies at scales features
3–4 vs. (black
1–2 asmarkers
the second inPCA
Figure 8) werediscussed
component also consistent
in the with
those inprevious
the X-ray section. The variations in the acoustic features (black markers in Figure 8) were also mainly
image textural features (red markers in Figure 8). When the materials
consistent with those when
contain cracks, especially in the oriented
X-ray image textural features
perpendicular (red wave
to the markers in Figure 8).direction,
propagation When the arrival
materials mainly contain cracks, especially when oriented perpendicular
times should be longer, the signals more attenuated (lower MAX and RMS features), and the energies to the wave propagation
direction, arrival times should be longer, the signals more attenuated (lower MAX and RMS features),
in lower frequency range (scales 3–4) should be higher compared to scales 1–2. Hence, those corridors
and the energies in lower frequency range (scales 3–4) should be higher compared to scales 1–2.
containing more
Hence, cracks
those should
corridors fall into more
containing the negative t1 region,
cracks should andthe
fall into those having
negative a higher
t1 region, andconcentration
those
of poreshaving
(and less cracks) should fall in the positive region. This was consistent
a higher concentration of pores (and less cracks) should fall in the positive region. This with thewas clustering
pattern consistent
shown inwith Figure 7. The X-ray
the clustering images
pattern shownof inselected corridors
Figure 7. The were also
X-ray images shown
of selected in Figure
corridors were9. Slices
5_1 andalso
7_1shown
both had in Figure 9. Slices
similar 5_1 and
t2 values but 7_1 both had similar
extremely t2 values
different but extremely
t1 scores compared different t1 scores5_6 and
to corridors
compared to corridors 5_6 and 7_6. Their X-ray images clearly showed that the former two corridors
7_6. Their X-ray images clearly showed that the former two corridors contained several transversal
contained several transversal cracks and some pores, whereas the latter two mostly contains pores,
cracks and some pores, whereas the latter two mostly contains pores, which were also larger.
which were also larger.

Figure 7. Score plot of the Partial Least Squares (PLS) model (t1–t2) between acoustic attenuation and
Figure 7. Score
X-ray plottextural
image of thefeatures.
Partial Observations
Least Squares are(PLS) model
identified (t1slice
by the –t2 )number
between(redacoustic
numbersattenuation
in Figure and
X-ray image textural
1) followed byfeatures. Observations
the corridor are Different
number (1–6). identified by the
colors areslice
usednumber (red numbers
to distinguish in Figure 1)
the different
followedcorridors
by the as
corridor
shown innumber (1–6).
the anode slice Different colors
image included in are used to distinguish the different corridors
the figure.
as shown in the anode slice image included in the figure.
The second component did not segregate the corridors as much as t1 did, as most of them
spanned the t2 range. Since it was orthogonal to the first component, the second component captured
Thea second
differentcomponent did not segregate
source of covariation between the
thecorridors as much
acoustic and as tfeatures.
textural 1 did, asThe
most of them
loading spanned
plot
the t2 range. Since it was orthogonal to the first component, the second component captured a different
Metals 2017, 7, 305 12 of 15

Metals 2017, 7, 305 12 of 15

source of covariation between the acoustic and textural features. The loading plot (Figure 8) shows that
all(Figure
features8)had
shows the that
sameallsign
features
in thehad the same sign
t2 component (all in the t2 component
positive), although the (all energies
positive),inalthough
scales 1–2 theand
most of the MAX and RMS features (except RMSD4) contributed less in this component. The samein
energies in scales 1–2 and most of the MAX and RMS features (except RMSD4) contributed less sign
forthis component.
these loadings The same
means
Metals 2017, 7, 305
that sign
they for
all these
variedloadings
together means
in the that
samethey all variedThis
direction. together
was in the same
consistent with
direction. This was consistent with the first component of the PCA model presented in the12previous of 15
the first component of the PCA model presented in the previous section, which was associated with
section, which
(Figure
was associated
8) shows density
withhad
that all features
variations
the volume
in material
same signof in voids.
density and/or
the t2 component
total volume
(all positive), although the
of voids.
variations in material and/or total Indeed, corridors containing more and
Indeed, corridors
energies containing
in scales 1–2 and more and
most would
of the larger
MAX voidsRMS(cracks and/or pores) wouldcontributed
lead to longer arrival
larger voids (cracks and/or pores) leadandto longer features
arrival(except
times,RMSD4)
and more energyless in
at scales 3–4
times,this
andcomponent.
more energy Theat scales
same 3–4
sign for(lower frequencies).
these loadings meansHence,
that theythey
all would project in
varied together in the
the positive
same t2
(lower frequencies). Hence, they would project in the positive t2 range compared with corridors
rangedirection.
compared with
This wascorridors
consistentcontaining
with the first fewer and smaller
component voids,
of the PCA which
model would in
presented falltheinprevious
the negative
containing
t2 area.
fewer
section,
The X-ray
and
which smaller
was
images
voids,
associated with
for corridors
which would
variations
3_3 and 15_3 in fall in the
material
had similar
negative
density and/or
t1 values,
t2but
area.
total The
volume
very
X-ray images
of voids.
different t2 scores
for
corridors
compared 3_3
Indeed, and
with 15_3
corridors had
corridors similar
containing
5_5 andmoret values,
1 and larger
24_5. but very different
voids (cracks
The images and/or
in Figure t scores
9 pores)
2
clearly compared
would
showed leadthat with
to longer corridors
arrival two
the former 5_5
andprojecting in the high t2 region contain more and larger voids (or defects), compared with thett22latter
24_5. The
times, images
and more in Figure
energy at 9
scalesclearly
3–4 showed
(lower that
frequencies). the former
Hence, theytwo projecting
would project in in
the the high
positive region
contain range
more compared
and largerwith corridors
voids (or containingcompared
defects), fewer and smaller
with voids,
the which
latter two would
fallingfallininthethe lower
negative t region.
two falling in the lower t2 region. 2
t2 area. The X-ray images for corridors 3_3 and 15_3 had similar t1 values, but very different t2 scores
compared with corridors 5_5 and 24_5. The images in Figure 9 clearly showed that the former two
projecting in the high t2 region contain more and larger voids (or defects), compared with the latter
two falling in the lower t2 region.

Figure 8. Loadings bi-plot (w2*-c1–w2*-c2) of the PLS model between acoustic attenuation signals and
Figure 8. Loadings bi-plot (w2 * -c1 –w2 * -c2 ) of the PLS model between acoustic attenuation signals and
X-rayFigure
image8.textural
Loadings features. (wThe
bi-plotThe features are identified by their names and by the scale at which
2*-c1–w2*-c2) of the PLS model between acoustic attenuation signals and
X-ray image textural features. features are identified by their names and by the scale at which they
they were
X-raycalculated (1–4).
image textural features. The features are identified by their names and by the scale at which
were calculated (1–4).
they were calculated (1–4).

Figure 9. X-ray images showing the distribution of cracks and pores in selected corridors.
Figure 9. X-ray images showing the distribution of cracks and pores in selected corridors.
Figure 9. X-ray images showing the distribution of cracks and pores in selected corridors.
Metals 2017, 7, 305 13 of 15

6. Conclusions
Baked carbon anodes are used in the Hall-Héroult aluminum reduction process to distribute the
electrical current throughout the cells, and to provide the source of carbon required by the electrolytic
reaction. However, poor quality anodes containing structural defects such as cracks, and a high
concentration of pores, can deteriorate the reduction cell performance by reducing energy efficiency
and increasing carbon consumption. The current quality control scheme based on collecting core
samples from a small proportion of the anode production and characterizing them in the laboratory
is not adequate for detecting pores and cracks in a rapid and non-destructive fashion. Research
efforts are currently being focused on developing new sensors for inspecting individual anode blocks
non-intrusively, at a speed compatible with industrial requirements.
Past work have demonstrated the potential of acousto-ultrasonic techniques (AU) to detect pores
and cracks in anode samples by using sequential excitation at different frequencies [6]. However, this
approach require long cycle times to collect all measurements. The present work aims at reducing cycle
times by 7 to 1 using a single frequency-modulated excitation, and showing that such an approach leads
to similar inspection results as those obtained with sequential excitation. A quantitative validation
of the inspection results is also proposed using X-ray images collected from each sample used in
this study.
A full size baked carbon anode was cut into 26 slices of equal thickness. A total of eight
slices distributed along the anode length and showing different internal structures were selected
for this work. After collecting X-ray images of the samples using a CT-Scan, they were excited by a
frequency-modulated wave at 6 different points along the slice width. The AU signal received after the
wave propagated through the material at each location (so-called corridors) was recorded. The Discrete
Wavelet Transform (DTW) was used to decompose the AU signals into different frequency bands,
and to perform texture analysis of the X-ray images. A number of scalar features were subsequently
calculated from the wavelet sub-signals and images. Multivariate latent variable methods, such as
PCA and PLS regression, were then used to cluster the corridors in each slice according to their internal
structure. A regression model was also built between the AU signals and the X-ray images in order to
relate variations in the acoustic signals with the presence of defects in the material.
The results showed that the textural features extracted from the X-ray images efficiently detected
the voids in the anode samples, and allowed the discrimination of pores from cracks, and captured the
severity of those defects. The PLS regression model built between the X-ray image textural and the AU
signal features for each corridor clearly showed that variations in the acoustic response of the samples
also allowed the detection and discrimination of both types of defects, and assessment of their severity.
Furthermore, the inspection results obtained with frequency-modulated excitation led to the same
conclusions as those obtained with sequential excitation [6], with a significantly shorter cycle time.
The proposed non-destructive acousto-ultrasonic inspection approach appeared to be very
promising for a real-time quality control of industrial scale pre-baked carbon anodes. Future work will
investigate the application of the inspection scheme to full size anodes collected from an industrial
manufacturing plant. A threshold will then need to be established for AU signals, and will serve
as a statistical control limit to determine whether an anode poses a higher risk of deteriorating the
reduction cell performance.

Acknowledgments: The authors would like to acknowledge the financial support of the Natural Sciences and
Engineering Research Council of Canada (NSERC), Fonds de Recherche du Québec—Nature et Technologies
(FRQNT), Alcoa and the Aluminum Research Centre—REGAL. The assistance provided by REGAL personnel,
Hugues Ferland and Guillaume Gauvin from Université Laval, in preparing the experimental set up is also
greatly acknowledged.
Author Contributions: Moez Ben Boubaker, Donald Picard, Jayson Tessier and Carl Duchesne conceived and
designed the experiments; Moez Ben Boubaker and Donald Picard performed the experiments; Moez Ben Boubaker
and Carl Duchesne analyzed the data with the support of Mario Fafard and Houshang Alamdari in the results
interpretation; Jayson Tessier supplied the materials; Moez Ben Boubaker, Donald Picard and Carl Duchesne
wrote the paper; Carl Duchesne, Houshang Alamdari and Mario Fafard revised the paper.
Metals 2017, 7, 305 14 of 15

Conflicts of Interest: The authors declare no conflict of interest.

References
1. Kocaefe, Y.; Kocaefe, D.; Bhattacharyay, D. Quality Control via Electrical Resistivity Measurement of
Industrial Anodes. In Light Metals; John Wiley & Sons: Hoboken, NJ, USA, 2015; pp. 1097–1102.
2. Andoh, M.A.; Kocaefe, Y.; Kocaefe, D.; Bhattacharyay, D.; Marceau, D.; Morais, B. Mesurement of the Electric
Current Distribution in An Anode. In Light Metals; John Wiley & Sons: Hoboken, NJ, USA, 2016; pp. 889–894.
3. Ziegler, D.P.; Secasan, J. Methods for Determining Green Electrode Electrical Resistivity and Methods for
Making Electrodes. U.S. Patent 9,416,458, 16 August 2016.
4. Chollier-Brym, M.J.; Laroche, D.; Alexandre, A.; Landry, M.; Simard, C.; Simard, L.; Ringuette, D.
New Method for Representative Measurement of Anode Electrical Resistance. In Light Metals; John Wiley &
Sons: Hoboken, NJ, USA, 2012; pp. 1299–1302.
5. Léonard, G.; Guérard, S.; Laroche, D.; Arnaud, J.C.; Gourmaud, S.; Gagnon, M.; Marie-Chollier, M.J.;
Perron, Y. Anode Electrical Resistance Measurements: Learning and Industrial On-line Measurement
Equipment Development. In Light Metals; John Wiley & Sons: Hoboken, NJ, USA, 2014; pp. 1269–1274.
6. Ben Boubaker, M.; Picard, D.; Duchesne, C.; Tessier, J.; Alamdari, H.; Fafard, M. The Potential of
Acousto-Ultrasonic Techniques for Inspection of Baked Carbon Anodes. Metals 2016, 6, 151. [CrossRef]
7. Suzuki, M.; Nazanishi, H.; Iwamoto, M. Relationship between acoustic emission characteristics and structural
factors of composite materials. In Proceedings of the Third Japan-U.S. Conference on Composite Materials,
Tokyo, Japan, 23–25 June 1986; pp. 631–638.
8. Ni, Q.-Q.; Jinen, E. Acoustic emission and fracture of carbon fiber reinforced thermosoftening plastic (CFRTP)
materials under monotonous tensile loading. Eng. Fract. Mech. 1993, 45, 611–625.
9. Ni, Q.-Q.; Jinen, E. Fracture behavior and acoustic emission of SFC. J. Soc. Mater. Sci. 1993, 42, 561–567.
[CrossRef]
10. Ni, Q.-Q.; Jinen, E. Acoustic emission technique in the single-filament-composite test. In Proceedings of the
First International Conference on Composite Engineering (ICCE/1), New Orleans, LA, USA, 28–31 August
1994; pp. 883–884.
11. Serrano, E.P.; Fabio, M.A. Application of the wavelet transform to acoustic emission signal processing.
IEEE Trans. Signal Process. 1996, 44, 1270–1275. [CrossRef]
12. Suzuki, H.; Kinjo, T.; Hayashi, Y.; Takemoto, M.; Ono, K. Wavelet transform of acoustic emission signals.
J. Acoust. Emiss. 1996, 14, 69–84.
13. Ni, Q.-Q.; Misada, Y. Analysis of AE signals by wavelet transform. J. Soc. Mater. Sci. 1998, 47, 305–311.
[CrossRef]
14. Cohen, L. Time-Frequency Analysis; Prentice Hall PTR: Englewood Cliffs, NJ, USA, 1995.
15. Kronland-Martinet, R.; Morlet, J.; Grossmann, A. Analysis of sound patterns through wavelet transforms.
Int. J. Pattern Recognit. Artif. Intell. 1987, 1, 273–302. [CrossRef]
16. Aussel, J.D.; Monchalin, J.P. Structure noise reduction and deconvolution of ultrasonic data using wavelet
decomposition (ultrasonic flaw detection). In Proceedings of the Ultrasonics Symposium, Montreal, QC,
Canada, 3–6 October 1989; pp. 1139–1144.
17. Loe, R.S.; Jung, K.; Anderson, K.; Abatzoglou, A.; Regan, J.; Arnold, H.; Lawton, W. Status report on wavelets
in signal detection and identification: Comparative processing and technology evaluation. In Proceedings of
the IEEE Sixth SP Workshop on Statistical Signal and Array Processing, Susono, Japan, 7–9 October 1992;
pp. 46–49.
18. Qi, G.; Barhorst, A.; Hashemi, J.; Kamala, G. Discrete wavelet decomposition of acoustic emission signals
from carbon-fiber reinforced composites. Compos. Sci. Technol. 1997, 57, 389–403. [CrossRef]
19. Qi, G. Wavelet-based AE characterization of composite materials. NDT E Int. 2000, 33, 133–145. [CrossRef]
20. Sandirasegaram, N.; English, R. Comparative analysis of feature extraction (2D FFT and wavelet) and
classification (Lp metric distances, MLP NN, and HNeT) algorithms for SAR imagery. In Proceedings of
the SPIE 5808, Algorithms for Synthetic Aperture Radar Imagery XII, Orlando, FL, USA, 14 June 2005;
pp. 314–325.
21. Mallat, S.G. A theory for multiresolution signal decomposition: The wavelet representation. IEEE Trans.
Pattern Anal. Mach. Intell. 1989, 11, 674–693. [CrossRef]
Metals 2017, 7, 305 15 of 15

22. Boespflug, X. Axial tomodensitometry—Relation between the Ct intensity and the density of the sample.
Can. J. Earth Sci. 1994, 31, 426–434. [CrossRef]
23. Picard, D.; Alamdari, H.; Ziegler, D.; Dumas, B.; Fafard, M. Characterization of a full-scale prebaked carbon
anode using X-ray computerized tomography. In Light Metals; John Wiley & Sons: Hoboken, NJ, USA, 2011;
pp. 973–978.
24. Borah, S.; Hines, E.L.; Bhuyan, M. Wavelet transform based image texture analysis for size estimation applied
to the sorting of tea granules. J. Food Eng. 2007, 79, 629–639. [CrossRef]
25. Bharati, M.H.; Liu, J.J.; MacGregor, J.F. Image texture analysis: Methods and comparisons. Chemom. Intell.
Lab. Syst. 2004, 72, 57–71. [CrossRef]
26. Geladi, P.; Grahn, H. Multivariate Image Analysis; John Wiley & Sons: Hoboken, NJ, USA, 1996.
27. Wold, S.; Esbensen, K.; Geladi, P. Principal component analysis. Chemom. Intell. Lab. Syst. 1987, 2, 37–52.
[CrossRef]
28. Höskuldsson, A. PLS regression methods. J. Chemom. 1988, 2, 211–228. [CrossRef]
29. Wold, S.; Sjöström, M.; Eriksson, L. PLS-regression: A basic tool of chemometrics. Chemom. Intell. Lab. Syst.
2001, 58, 109–130. [CrossRef]
30. Wold, S. Cross-validatory estimation of the number of components in factor and principal component
models. Technometrics 1978, 20, 397–405. [CrossRef]

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).