sensors

Article
Modular MA-XRF Scanner Development in the
Multi-Analytical Characterisation of a 17th Century Azulejo
from Portugal †
Sergio Augusto Barcellos Lins 1,2, * , Marta Manso 3 , Pedro Augusto Barcellos Lins 4 , Antonio Brunetti 5, *,
Armida Sodo 6 , Giovanni Ettore Gigante 1 , Andrea Fabbri 7 , Paolo Branchini 2,7 , Luca Tortora 2,6 and
Stefano Ridolfi 8

1 Dipartimento di Scienze di Base e Applicate per L’ingegneria, Università degli Studi di Roma “La Sapienza”,
Via Antonio Scarpa 14/16, 00161 Rome, Italy; [email protected]
2 Surface Analysis Laboratory Roma Tre—National Institute of Nuclear Physics—Roma Tre University,
Via della Vasca Navale 84, 00146 Rome, Italy; [email protected] (P.B.);
[email protected] (L.T.)
3 LIBPhys & VICARTE, Department of Physics & Department of Conservation and Restoration,
NOVA School of Science and Technology, 2829-516 Caparica, Portugal; [email protected]
4 Instituto de Química, Universidade Federal Fluminense, Rua Mario Santos Braga, 30-Centro,
Niterói, RJ 24020-140, Brazil; [email protected]
5 Istituto di Matematica e Fisica, Università degli Studi di Sassari, Via Vienna 2, 07100 Sassari, Italy
6 Department of Sciences, Via della Vasca Navale 84, 00146 Rome, Italy; [email protected]
7



INFN Sezione di Roma Tre, Via della Vasca Navale 84, 00146 Rome, Italy; [email protected]

 8 Ars Mensurae, Via Vincenzo Comparini 101, 00188 Rome, Italy; [email protected]
Citation: Barcellos Lins, S.A.; Manso, * Correspondence: [email protected] (S.A.B.L.); [email protected] (A.B.)
† This paper is an extended version of our paper published in IMEKO TC-4 MetroArchaeo 2020, Trento, Italy,
M.; Barcellos Lins, P.A.; Brunetti, A.;
22–24 October 2020.
Sodo, A.; Gigante, G.E.; Fabbri, A.;
Branchini, P.; Tortora, L.; Ridolfi, S.
Abstract: A modular X-ray scanning system was developed, to fill in the gap between portable
Modular MA-XRF Scanner
Development in the Multi-Analytical
instruments (with a limited analytical area) and mobile instruments (with large analytical areas, and
Characterisation of a 17th Century sometimes bulky and difficult to transport). The scanner has been compared to a commercial tabletop
Azulejo from Portugal. Sensors 2021, instrument, by analysing a Portuguese tile (azulejo) from the 17th century. Complementary techniques
21, 1913. https://doi.org/10.3390/ were used to achieve a throughout characterisation of the sample in a complete non-destructive
s21051913 approach. The complexity of the acquired X-ray fluorescence (XRF) spectra, due to inherent sample
stratigraphy, has been resolved using Monte Carlo simulations, and Raman spectroscopy, as the most
Academic Editor: Giovanni Leucci suitable technique to complement the analysis of azulejos colours, yielding satisfactory results. The
colouring agents were identified as cobalt blue and a Zn-modified Naples-yellow. The stratigra-
Received: 10 February 2021
phy of the area under study was partially modelled with Monte Carlo simulations. The scanners
Accepted: 5 March 2021
performance has been compared by evaluating the images outputs and the global spectrum.
Published: 9 March 2021

Keywords: MA-XRF; Monte Carlo simulations; azulejos; Raman spectroscopy

Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
1. Introduction
Decorative ceramic glazed tiles (azulejos) are an intrinsic part of Portuguese culture,
being widely used in the country to decorate the most varying surfaces since the 15th
Copyright: © 2021 by the authors.
century until today. Their local use is believed to have started around the 13th century
Licensee MDPI, Basel, Switzerland.
through Moorish influence, while the majolica technique arrived in Portugal only later,
This article is an open access article
around the 16th century, with the establishment of local workshops [1,2].
distributed under the terms and The azulejos are mainly composed of a ceramic biscuit and glaze, the latter comprising
conditions of the Creative Commons a wide gamut of colours, which can act as a fingerprint of the azulejo, allowing the identi-
Attribution (CC BY) license (https:// fication of its origin, period, and manufacture technique [1,3–5]. Their stratigraphy and
creativecommons.org/licenses/by/ composition can be analysed by sampling and cross-sectioning the artefact, an approach
4.0/). not always feasible or desired. An alternative way to avoid a destructive approach is to

Sensors 2021, 21, 1913. https://doi.org/10.3390/s21051913 https://www.mdpi.com/journal/sensors

Sensors 2021, 21, 1913 2 of 13

perform Monte Carlo simulations of X-ray fluorescence spectra. By iteratively simulating

the experimental data, it is possible to model the stratigraphy and composition of the
irradiated sample up to a satisfactory degree of accuracy [6–8].
When it comes to a complete non-destructive analysis, the glaze and pigments
chemical composition can be investigated by X-ray fluorescence (XRF) and Raman spec-
troscopy [1,4,5]. The XRF punctual analysis, in particular, yields local information, which
does not always represent the sample as a whole. In this case, it is convenient to map the
artefacts’ surface, gaining information on the spatial distribution of chemical species.
The use of X-ray fluorescence scanners in heritage sciences is becoming a common
practice, thanks to the spatial information on chemical elements being more crucial for the
analysis of painted surfaces, and the growing popularization and accessibility of these in-
struments [9–13]. However, it is often the case where these artworks cannot be transported
into laboratories or synchrotron facilities due to their valuable nature [14]. This has led to
a spike in the development of transportable and, more recently, portable systems [15–17].
Yet, the design choice of developing a portable or mobile X-ray fluorescence scanner comes
with crucial limitations. A portable unit offers a limited scanning area, while a mobile one
provides a much larger scanning area, at the cost of sacrificing its portability.
When developing an in-house and accessible scanning system, other parameters, such
as pricing, must be taken into consideration. Most of the systems’ cost is related to the X-ray
electronics. This cost can be reduced if low-power electronics and simple focusing optics are
used, with the drawback often resulting in longer dwell-times and lower resolutions [18,19].
Furthermore, as system portability and maximum scanning area are competing parameters,
difficult choices, apart from budget, must be made when considering developing one single
scanning system, sacrificing either one or another parameter.
In this scope, combining cost-effectiveness and versatility, a modular scanning unit
was developed in a joint effort between Ars Mensurae, Istituto Nazionale di Fisica Nucleare—
Roma TRE and Sapienza University. The system was put to test against a commercial
tabletop µ-XRF scanner. The data collected with both instruments was used, combined
with other analytical techniques, to fully characterize a Portuguese azulejo belonging to the
17th century.

2. Materials and Methods

A Portuguese tile (azulejo) from approximately the 17th century was investigated with
the proposed Macro-X-Ray Fluorescence (MA-XRF) scanning system and complementary
techniques. The azulejo is shown in Figure 1, where four colours can be identified—yellow
(A), white (B), orange (C), and blue (D). Area 1 represents the region investigated through
MA-XRF scanning.
Monte Carlo simulations of XRF spectra acquired in different spots of the tile were
performed to better elucidate the stratigraphy and composition of the yellow, white and
blue coloured areas. Furthermore, µ-Raman analysis of several other spots were per-
formed, serving as a complementary technique to better understand the nature of the
colouring agents.

2.1. MA-XRF Scanning

Elemental distribution maps from a ≈25 cm2 portion of the azulejo were acquired with
the Modular MA-XRF Scanner. The results were compared with those obtained with a
commercial, tabletop scanner from BRUKER® GmbH, Germany. The analysis conditions
and system setups, as well as a description of the Modular Scanner, are presented in the
following subsections.
Sensors 2021, 21, 1913 3 of 13

Figure 1. Sample and Macro-X-Ray Fluorescence (MA-XRF) scanned area (1). The colours identified
are: yellow (A), white (B), orange (C) and blue (D).

2.1.1. System Development

Figure 2 shows a schematic drawing of the translation stages designed for the scanning
system described herein. To overcome one of the major disadvantages of portable systems,
two translation stages were developed, one being small and portable, capable of scanning
areas up to 20 × 20 cm2 , and a larger, mobile one.

Figure 2. Translation stages: portable version (A) and mobile version (B).

Multiple X-rays detectors can be fitted in the scanning head to reach shorter dwell-
times if needed. The X-ray tube can be exchanged, as the head threading can fit any of
Moxtek’s® analog-handheld tubes. The beam is focused with an exchangeable aluminium
collimator. The current prototype iteration uses up to 3 AMPTEK® X-123 Silicon-Drift
Detectors (SDD), which are low power and lightweight. The detectors are configured
identically, so acquisition can be synchronized and the sum spectrum from all detectors
viewed in live time.
In addition to the X-rays electronics, the x-y translation stage can be exchanged as
well. This allows one to achieve quick assembly times (ca. 25 min) as well as provides an
extra degree of versatility. X-rays detectors and translation stage(s) are controlled by the
same acquisition software, developed in the LabVIEW® platform. The tube is controlled
externally, not integrated in the controlling software, enforcing the system’s modularity.
Sensors 2021, 21, 1913 4 of 13

In sum, the system can be configured to suit each necessity. The scanning head
(with only one detector), cables, small stage and all related peripherals can comfortably fit
inside a standard airplane cabin trolley. The portable version is described in more detail
elsewhere [20].

2.1.2. Data Acquisition

The areas scanned by both the Modular Scanner and the M4 Tornado are highlighted
in Figure 1. Although M4 Tornado can provide considerably shorter dwell-times, thanks
to its higher power X-ray tube, both analyses were performed under the most similar
conditions possible, to ascertain whether the results are comparable. Scanned areas were
5.3 × 5.0 cm2 in size each and were acquired with a tube voltage of 35 kV and with an
applied current of 29 µA. The areas scanned with each instrument were slightly offset from
one another and were cropped to improve the comparison and overlay between them.
Dwell-times were set to 500 ms for the M4 Tornado and 470 ms for the Modular
Scanner, with an overall stage speed of approximately 2 mm/s. The M4 Tornado scanner
is equipped with a Rh target X-ray tube, while the Modular Scanner had an Ag target
Moxtek® tube attached. The Modular Scanner was assembled with the mobile stage and
two X-123 SDD AMPTEK® detectors.
Overall scanning time was about 20 min for each instrument, with both acquisitions
being performed in a continuous scan fashion. Elemental distribution maps were generated
by different software. M4 Tornado images were obtained with Bruker ESPRIT proprietary
software. Data obtained with the Modular Scanner were processed with an in-house
developed software (XISMuS) version 1.3.2 [21].

2.2. Monte Carlo Simulations

The XRF spectra simulations were carried with a modified version of the XRMC
package [22], version 6.4.1., a Monte Carlo algorithm based on the xraylib [23] database.
This XRMC version is capable of simulating rough surfaces as well as up to any order of
photoelectric interaction. However, for the present purpose, a threshold was set at the 3rd
order, since errors in the atomic parameters accumulate as the interaction order increases,
and results may lose significance.
Experimental data were collected with a portable XRF spectrometer, from 3 spots
within the sample surface, representing the white, yellow, and blue colours. For the blue
colour, two spectra were collected, one for the light shade of blue and one for the darker
shade, in the right-most part of area 1 (Figure 1). The spectrometer used was composed
of a Mini-X tube, from AMPTEK® with a silver anode, collimated to 1 mm and an X-123
SDD X-ray detector, also from AMPTEK® . The system geometry was set as the following:
detector parallel to the sample’s normal, 3 cm distant and uncollimated; tube at 45 degree
from the sample’s normal and 2 cm distant. The tube operated with a voltage of 40 kV and
an applied current of 15 µA.
Each simulation result is obtained iteratively, with user intervention at each iteration
step. The simulation starts with an initial guess, essentially based on the peaks visible in
the measured spectrum and, when possible, on the typical composition of the sample as
reported in literature as well. The X-Ray excitation spectrum and geometrical factors of
the portable XRF instrument have been carefully modeled using several measurements of
reference samples.
The azulejo has been modeled as a set of layers, each of them identified by six flat
planes (called quadrics in terms of XRMC parameters). The simulated and measured
spectra are compared at each iteration, and, if differences are detected, the simulation
parameters are adjusted accordingly. These are all significative parameters required to
perform a simulation. This protocol has been previously and successfully applied to other
archaeological materials, and is described in more detail elsewhere [7,8,24].
Sensors 2021, 21, 1913 5 of 13

2.3. µ-Raman Analysis

Raman spectra from 4 different spots, covering all the detected colours, were collected
with the InVia™ µ-Raman spectrometer from RENISHAW® , UK, equipped with a Leica
DM2700 M confocal microscope. The spots were analysed with two different solid-state
diode laser sources, one at 532 nm and the second at 785 nm, selected according to need.
The former source operated at 120 mW power while the latter operated at 250 mW. Neutral
filters were used to decrease the power at the sample, taking into account its nature and
composition thus avoiding degradation processes. A set of long working distance objectives
(namely 20× and 50×) was used, focusing the excitation beam down to few microns.
To achieve enough statistics, 3 to 5 scans, with an integration time of 5 s were acquired.
The data were collected with the instrument’s proprietary software and then exported as
ASCII files. The output spectra were processed with a graphic analysis software.

3. Results and Discussion

3.1. Scanner Performance
The in-house developed instrument had its performance compared to a state-of-the-art
one, and the results obtained can be seen in Figure 3. It is important to highlight that
the optics chosen can play a significant role when it comes to final resolution, as the
irradiation spot can vary widely (from few millimetres to few micrometres). In the case
of our instrument, an aluminium collimator of 2 mm diameter was used, while the M4
Tornado had a polycapillary lens focusing the beam (at 20 µm for Mo-Kα energy and 38 µm
for Mo-Lα energy). This ensued slightly more blurred images with the Modular Scanner
in respect to the M4 Tornado. Significant differences can be observed in all images of As,
Sn, Sb, and Pb elemental distribution maps. At first, it is plausible to assume this may be
due to the M4 Tornado detector’s resolution and number of channels, the latter 4 times
greater than that set at the X-123 SDD detectors in the Modular Scanner, conceiving some
significant superior resolving power. Moreover, the narrower beam provided by the M4
Tornado instrument, even though provides an inferior count rate (Supplementary Material
Figure S2), allows the collection of more localized information.

Figure 3. Elemental distribution maps obtained with the Modular Scanner and the M4 Tornado.
Scale is 10 millimetres.
Sensors 2021, 21, 1913 6 of 13

The elemental distribution maps were obtained through different data processing
routines, as different software were used to process the raw data from each instrument.
Because of that, a direct comparison between each image pair was performed. Since
few images, from both M4 Tornado and the modular scanner, were too noisy, blurred or
with a low dynamic range, they had to be first filtered to render this operation possible,
otherwise the mean-squared error test would yield biased information. Furthermore,
each pixel represents a conversion from the images’ respective element net-peak area to a
greyscale (0–255) intensity value, requiring that few images were normalized beforehand.
Finally, each image pair had its histogram corrected. This was performed by applying
a histogram-matching algorithm followed by a linear contrast stretching one. Rotation
corrections were dismissed, as the images already present a good agreement in this regard.
The mean-squared error (MSE) and structural similarity index (SSIM) scored for each pair
is given in Table 1.

Table 1. Image scores between the Modular Scanner and M4 Tornado images.

Kα Kβ Lα Mα
Element MSE SSIM MSE SSIM MSE SSIM MSE SSIM
Fe 539.30 0.71 - - - - - -
Co 556.82 0.68 - - - - - -
Ni 678.35 0.44 - - - - - -
Zn 781.04 0.40 - - - - - -
As - - 1116.47 0.27 - - - -
Sn 1673.41 0.11 - - 1957.48 0.11 - -
Sb 2194.59 0.21 - - 1167.72 0.50 - -
Pb - - - - 146.66 0.14 2883.70 0.12

SSIM algorithm measures the quality of an image in comparison to a reference one,

working closer to the human visual system instead of the mathematically defined methods
(as MSE) [25,26]. SSIM scores rate how similar two images are, 0 meaning entirely different
images, and 1 meaning the two images are identical. As for MSE values, lower values
mean there is theoretically less difference between the two images. Equation (1) shows
how the MSE is calculated, where n is the total number of pixels, and y0 and y1 are the
images ith-pixels. Detailed results can be found in the Supplementary Material Figure S1.
n
1
MSE =
n ∑ ( y0 − y1 )2 . (1)
i =1

Arsenic images are considerably noisy, and a comparison between them would clearly
yield unsatisfactory results for both MSE and SSIM, despite matching their histograms.
The concomitant presence of Pb and As makes it hard to obtain a clear image from As-Kα
line, as a significant overlapping with Pb-Lα line occurs. The obvious option is to create an
image from As-K β line, which is clearly resolved but with significant less counts, due to
both its low content in the sample and the inherent lower probability of K-beta emissions
over K-alpha.
Most images present a low SSIM score (from 0.11 to 0.26), in particular Sn and Pb
images. The structural similarity test, in fact, tends to yield a low score when comparing
blurred images. For example, a same image, compared with its highly interpolated and en-
larged version, will score poorly on the SSIM test [25] (Supplementary Material Figure S1l).
For these images, the MSE works better as a comparison parameter.
From all elemental distribution maps obtained, the most discrepant are Sn-Kα , Sb-Kα ,
and Pb-Mα . The former being close to unreadable regarding the results from both instru-
ments. M4 Tornado’s Sn- and Sb-Kα images are entirely blurred, and the Modular Scanner
Pb-Mα is completely noisy. When taking into consideration that the Modular Scanner was
capable of producing an image for Sb-Kα and a better image for Pb-Lα , while M4 Tornado
Sensors 2021, 21, 1913 7 of 13

produced better images for Sn- and Sb-Lα , and Pb-Mα lines, one can clearly see where the
optimal working condition of each instrument lies.
The use of a polycapillary lens provided a much narrower beam, at the cost of sacri-
ficing resolving power at higher energies, a phenomenon better described elsewhere [27].
Furthermore, Sn and Pb are expected to be present in higher quantities in the vitreous base
layer, from which the signal is considerably attenuated by the superimposing pigments
strata. A less intense excitation beam would highlight this factor even more, as it can be
seen in Figure 4. Even though M4 Tornado sum spectrum presents narrower peaks for Sn-
and Sb-Kα lines, with apparently much less background contribution at the higher energy
end of the spectrum, the count-rate observed was insufficient to produce clear images.

Figure 4. Sum spectra of the scanned regions.

Regarding the lower energy end of the spectra, M4 Tornado outperforms the Modular
Scanner, providing a clearer image for Sb-Lα and overall better images for Sn-Lα and
Pb-Mα . Figure 4 shows that indeed the M4 Tornado provides a better energy resolution,
signal-to-noise ratio and count-rate at the very low end of the spectrum. From mid to high
energy range (≈5 to 30 KeV), the use of a collimator provided better results, with higher
counts, and, therefore, better images; for example, those of Sb-Kα and Pb-Lα .
For Sn and Sb, the use of polycapillary lenses can justify the loss of excitation intensity
in the mid-to-high energy interval to gain on resolution, yielding more defined images at
the lower energy range (L-lines). As for the remaining images, advantages provided by a
smaller beam size, at the given analysis conditions, were not fully exploited. Co and Fe
images, for example, have all fine details visible, for both instruments. Therefore, for larger
and faster scans, where the beam size if often wider (≈1 mm), the costs trade-off between
having an instrument with collimator optics rather than polycapillary lenses can be fully
justified. Furthermore, collimators dismiss the need of a z-axis focusing system, as the
output beam is less divergent and the irradiation spot is almost independent from the tube
distance [27].
In any case, regardless of the optics chosen, the elemental distribution maps provided
by the MA-XRF technique already yielded some relevant information from which it is
possible to layout a multi-analytical strategy. The azulejo in question is clearly a majolica
style tile, where coloured oxides were applied over a white lead-tin glaze. The layers
disposition can first be assumed by interpreting the K- and L-lines or L- and M-lines of a
same element, as the case of Sn, Sb, and Pb.
To further understand the pigments and technique used in its manufacture, comple-
mentary techniques as Raman and Monte Carlo simulations were used. The different
pigments and layers are analysed individually in the following subsections.
Sensors 2021, 21, 1913 8 of 13

3.2. White Glaze

The white, vitreous phase, that characterises the majolica technique, is commonly
composed of Pb oxides, and Sn and Si dioxides [1]. Lead oxide is usually used to lower
the siliceous matrix melting point, while tin dioxide is used as an opacifying agent. The
resulting mixture (phase) can be porous and have a considerable amount of inclusions [1,5].
The elemental distribution maps, shown in Figure 3, indicate that the white glaze,
present in the azulejo herein analysed, has lead and tin in its composition, in accordance
to other white glazes observed in azulejos from the same period. Furthermore, with
Monte Carlo simulations, it was possible to estimate this layer’s thickness, which varies
from 280 to 350 µm, also compatible with previously reported tiles [1]. Raman spectra
confirmed the presence of cassiterite (SnO2 ), evidenced by bands at 635 and 775 cm−1 , a
compound normally used as an opacifying agent. Simulation spectra can be found in the
Supplementary Material.

3.3. Blue Colour

Closely observing the elemental maps in Figure 3, it is possible to notice a strong
correspondence between cobalt, nickel, iron, and arsenic. The latter being an indication
of the cobalt oxide origin, and iron, having a correlation that is predominant in the left-
most part and around the central S-shaped region. This quaternary combination can be
directly related to zaffre, imported by Portugal from Germany in the beginning of the 16th
century [1]. The information from the elemental maps alone gives a hint on the pigment
used to obtain the blue colour, although a clear distinction between the light and dark
shades was not achieved.
Figure 5 shows the µ-Raman spectra obtained for the blue glazed regions, which
seems similar at first sight, besides representing different intensities of blue. One is a dark,
intense blue, and the other, a lighter blue with glassy appearance. This difference could be
explained due to the presence of a Raman band peaking at 824 cm−1 that allows a clear
identification of a cobalt blue pigment: cobalt silicate (CoSiO4 ), olivine type, which has
been previously reported by de Waal [28] and, recently, Ferreira [29].
The blue colour was traditionally obtained from cobalt oxides (CoOx ) which dissolve
in the glassy phase (SiO2 ). Co ions interact with the Si-O network and create new Si-
O-Co bonds by replacing Si in the glass structure. In the 1st half of 17th century, this
technique was done by dissolving cobalt ore in low quantities into the glaze, which does
not necessarily lead to the precipitation of the silicate. As expected, the 824 cm−1 signature
band is not observed for blue pigments in ceramics of this period [30]. On the other hand,
in the 2nd half of the 17th century, this signal is detected, which leads to the assumption
that the glazing technique was done using higher temperatures [28,31].

Figure 5. Recorded µ-Raman spectra for blue glazed regions: (A) vitreous blue region, (B) darker
blue region; (B) λexcitation = 532 nm.
Sensors 2021, 21, 1913 9 of 13

The difference between the blue glazed regions is probably due to the content of Co
in the structure. In the vitreous blue region (Figure 5A) a lighter blue is observed, due to
a lower presence of the cobalt pigment. So, there is a predominance of the glassy silicate
structure, which can be detected by the Si–O stretching and bending phases (envelopes at
980 and 465 cm−1 ). In the darker blue region (Figure 5B), as expected, the 824 cm−1 is far
more noticeable, as observed by Pereira et al. [31]. The difference in the relative amount of
cobalt, for the bright and dark shades of blue, can also be observed in the elemental maps
shown in Figure 3, where a more intense signal for this element is noticed in the rightmost
part, together with arsenic.
Furthermore, Monte Carlo simulations of XRF spectra obtained from the blue colour,
yielded a thickness of about 70 µm for both shades, light and dark. The difference relied
in the chemical composition, where the dark blue presented from 2 to 4 times more iron,
nickel, cobalt and arsenic. This supports the other information obtained, showing that the
dark blue shade has a higher concentration of cobalt and associated elements.

3.4. Yellow Colour

The elemental distribution maps show the presence of Zn, Pb, and Sb in the yellow
regions (Figure 3), suggesting the use of Naples Yellow (Pb2 Sb2 O7 ) or a Zn-modified version
of this pigment. To verify this hypothesis, the µ-Raman spectra (Figure 6) were used. The
yellow-region spectra do not exactly match with the reference spectra for Naples Yellow
(peaks at 136, 328, 453 and 510 cm−1 ), indicating indeed the possibility of a ternary oxide
structure. Several studies [32–34] draw attention for these ternary oxides, as variations from
the standard Naples Yellow by the addition of either Zn and/or Sn, replacing some of the
Sb in the pigments structure [33] and thus creating Sn-Pb-Sb and/or Zn-Pb-Sb pyrochlores.

Figure 6. Recorded Raman spectra for a yellow glazed region (λexcitation = 532 nm).

The very strong band at 130 cm−1 is reported for the Pb-O lattice vibration at low
wavenumbers (120–139 cm−1 ), and should not be used as a good discrimination band for
ternary oxides. Furthermore, it varies significantly even for unmodified Naples Yellow and
different firing temperatures for the pigment [34]. It is worth noting that for Zn modified
Naples Yellow, this band usually appears at 145 cm−1 , while for the Sn modified, it shows
a doublet at 125 and 142 cm−1 [33].
In unmodified Naples Yellow there is a strong band at 513 cm−1 for the SbO6 octahedra
symmetric elongation. This band is much more useful for discrimination and character-
ization of ternary oxides, since it collapses and shows another band at ca. 450 cm−1 for
modified pigments [1]. Even though the collected spectrum shows a band at ca. 460 cm−1 ,
it is worth noting that this band can present variations with the type of ternary oxide
and composition. Nonetheless, is can still be used as a good indicator of the presence of
ternary oxides.
Sensors 2021, 21, 1913 10 of 13

Furthermore, accordingly to Sandalinas et al. [34], for Sn-Pb-Sb pyrochlores there

should be a band at ca. 640 cm−1 , which could not be detect in the obtained spectra for the
yellow region. This further reinforces the possible presence of a Zn-modified Naples Yellow.
Finally, Monte Carlo simulations revealed the presence of a multi-layered structure
in the yellow region, demonstrating that the yellow pigment was painted on top of the
white glaze, according to the majolica technique, and with a thickness of about 50 µm.
The composition of the yellow strum, in the simulations, revealed a high content of lead
and antimony, as expected and demonstrated by both the Raman spectra and elemental
distribution maps. Zinc, on the other hand, was found in low percentages within the
simulations, but present nonetheless.

3.5. Orange Colour

When observing Pb-Lα and Fe elemental distribution maps (Figure 3), one can notice
an absence of correlation between these two elements in what concerns the orange region
(resembling an S-shape). However, in an opposite way, Zn and Sb present a correlation
with the orange region and the Pb-Mα image, analogously to the yellow region. From the
Pb images, it is possible to assume that lead oxides are also contained within the orange
matrix. This indicates an intentional addition of iron oxides (possibly rust [35]) to the
ternary yellow pigment in order to obtain the observed orange hue.
Analysing the orange colour under a microscope, the pigment reveals itself more
similar to the yellow one indeed, but rougher, with the presence of dark spots, and without
the characteristic shiny and transparent layer present on top (Figure 7). This last stratum
(coperta) is a very thin layer of powdered glass sprinkled on top of ceramics in order to
conceive a shiny and smooth aspect to the final product [35].

Figure 7. Micrographs from the yellow, blue and orange colours. Scale bar is 2 mm wide.

To further understand the pigments’ nature, µ-Raman spectra (Figure 8) were analysed.
They appear to be very akin to the spectra obtained from the yellow region, showing
similar bands and peaks (at ca. 131, 302 and 512 cm−1 ) and indicating the use of a mixture
obtained from the modified Naples Yellow and other components, as previously seen in
the yellow colour.
The recorded spectrum for a dark grain (Figure 7) observed within the orange colour
composition (Figure 8) shows very visible bands at ca. 293, 410 and 610 cm−1 , which
matches with α-hematite (Fe2 O3 ) reference spectrum [30]. This can be an indicative that
the modified Naples Yellow was indeed mixed with hematite to obtain the desired orange
colour. This hypothesis can be partially supported by the strong presence of Fe in the XRF
spectra for the orange region and previous studies on orange colours used on azulejos from
the same period [1,4].
Monte Carlo simulations did not yield satisfactory results for this particular pigment,
as the stratigraphy appears to be relatively over-complex and there was not much conclu-
sive information (possible layering structure and layers composition) to use as an initial
guess to feed the model.
Sensors 2021, 21, 1913 11 of 13

Figure 8. Comparison of recorded Raman spectra obtained from the orange region: orange grain,
yellow matrix, and dark grain. The λexcitation for the dark grain spectrum was of 785 nm, while for
the remaining spectra, 532 nm.

4. Conclusions
The in-house developed modular MA-XRF scanner proposed had its performance
compared to a commercial instrument. The results obtained from both instruments were
compared and discussed. A clear difference in resolution, in the lower energy end of the
spectra, probably thanks to the more expensive, sub-mm X-ray optics and X-ray detector,
used in the commercial system could be observed. However, for general purposes, larger
and faster scans, the use of cost-effective optics could be justified.
Moreover, not only did the proposed instrumentation performance proved satisfactory
but also it demonstrated to be a good alternative to expensive and bulky commercial
instrumentation. The modular scanner can be faced as a fully portable unit or a mobile
instrument, performing similarly to fixed instruments under similar analysis conditions.
Elemental distribution maps obtained from both scanners were combined with two
complementary analytical techniques, to thoroughly investigate a 17th century Portuguese
azulejo. The overall stratigraphy of the sample, as well as the identification of most colour-
ing agents was possible with MA-XRF data alone. Raman analysis was fundamental in
understanding the differences in the shades of blue found in the sample as well as confirm-
ing the yellow pigment hypothesis. Monte Carlo simulations proved to be a powerful tool,
determining the stratigraphy, composition and thickness of almost all pictorial layers.
In summary, the pigments identified in the azulejo were: (a) cobalt blue (obtained from
Zaffre) in varying concentrations to obtain different shades of blue, (b) Zn-modified Naples
Yellow, and, (c) a combination of hematite and Zn-modified Naples Yellow, to obtain the
orange colour. The white glaze was found to be a silica (SiO2 ) matrix, modified with the
use of lead oxide (flux) and tin dioxide (opacifier).

Supplementary Materials: The following are available online at https://www.mdpi.com/1424-822

0/21/5/1913/s1, Figure S1: Detailed SSIM and MSE comparisons, Figure S2: Full sum spectrum of
scanned region, Figures S3–S6: Monte Carlo simulated and experimental spectra comparisons, Table
S1: Monte Carlo layers final composition, Video S1: MA-XRF scanning 3D render.
Author Contributions: Conceptualization, S.A.B.L., M.M. and G.E.G.; methodology, S.A.B.L.; soft-
ware, S.A.B.L., A.F. and A.B.; validation, M.M., G.E.G. and A.B.; formal analysis, P.A.B.L. and A.B.;
investigation, S.R., A.S. and A.B.; resources, S.R., M.M., A.B., P.B., L.T. and A.S.; data curation,
S.A.B.L.; writing–original draft preparation, S.A.B.L. and P.A.B.L.; writing–review and editing, L.T.,
M.M. and G.E.G.; visualization, S.A.B.L., P.A.B.L. and M.M.; supervision, G.E.G.; project administra-
tion, P.B., S.R., G.E.G. and L.T.; funding acquisition, M.M., G.E.G., A.B., P.B., A.F. and L.T. All authors
have read and agreed to the published version of the manuscript.
Sensors 2021, 21, 1913 12 of 13

Funding: This project has received funding from the European Union’s Horizon 2020 research
and innovation programme under the Marie Skłodowska-Curie grant agreement No. 766311, from
Regione Lazio, under the Project “MUSA” n. B86C17000280002, of the Excellence Centre at the Lazio
Technological District for Cultural Heritage (DTC), from the Regione Autonoma di Sardegna, under
project number CUP:J81G17000140002, and from Fundação para a Ciência e a Tecnologia, grants
UID/FIS/04559/2020 and UIDB/00729/2020.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The data presented in this study are available upon request to the authors.
Acknowledgments: The authors would like to thank Maria Luísa Carvalho for kindly providing the
azulejo sample for investigation, and Alexandre Pais for the helpful comments and for historically
dating the sample.
Conflicts of Interest: Author Stefano Ridolfi was employed by the company Ars Mensurae srl. The
remaining authors declare that the research was conducted in the absence of any commercial or
financial relationships that could be construed as a potential conflict of interest.

References
1. Coentro, S.; Mimoso, J.M.; Lima, A.M.; Silva, A.S.; Pais, A.N.; Muralha, V.S. Multi-analytical identification of pigments and
pigment mixtures used in 17th century Portuguese azulejos. J. Eur. Ceram. Soc. 2012, 32, 37–48. [CrossRef]
2. Leal, A.S.; Alves, L.C.; Coentro, S.; Pereira, S.; Relvas, C.; Ferreira, T.; Mirão, J.; Fernandes, L.; Muralha, V.S. Caracterização
química, física e mineralógica da colecção de azulejos hispano-mouriscos do Museu de Lisboa—Teatro Romano. Conservar
Património 2018, 29, 25–39. [CrossRef]
3. Barcellos Lins, S.A.; Manso, M.; Gigante, G.E.; Cesareo, R.; Tortora, L.; Branchini, P.; Ridolfi, S. Modular MA-XRF scanner
potentialities and further advances. In Proceedings of the IMEKO International Conference on Metrology for Archaeology and
Cultural Heritage, Trento, Italy, 22–24 October 2020; pp. 496–500.
4. Mortari, C.; Nobre Pais, A.; Esteves, L.; Gago da Câmara, A.; Carvalho, M.L.; Manso, M. Raman and X-ray fluorescence glaze
characterisation of Maria Keil’s decorative tile panels. J. Raman Spectrosc. 2021, 52, 59–70. [CrossRef]
5. Mimoso, J.M.; Pais, A.; Ferreira, M.; Esteves, M.d.L.; Pereira, S.R.M.; Antunes, M.A.; Valona, R.; Cardoso, A.M.; Candeias,
A. Instrumental Study of the 16th Century Azulejo Panel Decorating a Public Fountain in Alcacer do Sal-Portugal; LABORATÓRIO
NACIONAL DE ENGENHARIA CIVIL, I.P. LNEC: Lisbon, Portugal, 2019; pp. 19–34.
6. Barcellos Lins, S.A.; Gigante, G.E.; Cesareo, R.; Ridolfi, S.; Brunetti, A. Testing the Accuracy of the Calculation of Gold Leaf
Thickness by MC Simulations and MA-XRF Scanning. Appl. Sci. 2020, 10, 3582. [CrossRef]
7. Brunetti, A.; Golosio, B.; Schoonjans, T.; Oliva, P. Use of Monte Carlo simulations for cultural heritage X-ray fluorescence analysis.
Spectrochim. Acta Part B At. Spectrosc. 2015, 108, 15–20. [CrossRef]
8. Brunetti, A.; Fabian, J.; La Torre, C.W.; Schiavon, N. A combined XRF/Monte Carlo simulation study of multilayered Peruvian
metal artifacts from the tomb of the Priestess of Chornancap. Appl. Phys. A 2016, 122, 571. [CrossRef]
9. Ravaud, E.; Pichon, L.; Laval, E.; Gonzalez, V.; Eveno, M.; Calligaro, T. Development of a versatile XRF scanner for the elemental
imaging of paintworks. Appl. Phys. A Mater. Sci. Process. 2016, 122, 1–7. [CrossRef]
10. Dik, J.; Janssens, K.; Van Der Snickt, G.; Van Der Loeff, L.; Rickers, K.; Cotte, M. Visualization of a lost painting by Vincent van
Gogh using synchrotron radiation based X-ray fluorescence elemental mapping. Anal. Chem. 2008, 80, 6436–6442. [CrossRef]
11. Nervo, M.; Romano, F.P.; Caliri, C.; Piccirillo, A.; Triolo, P.; Demonte, D.; Gatti, A.; Vergallo, E.; Cardinali, M.; Ferrero, M.
“Costruzione del viadotto”: MA-XRF in the pictorial executive technique of Agostino Bosia. X-ray Spectrom. 2020. [CrossRef]
12. Turner, N.K.; Patterson, C.S.; MacLennan, D.K.; Trentelman, K. Visualizing underdrawings in medieval manuscript illuminations
with macro-X-ray fluorescence scanning. X-ray Spectrom. 2019, 48, 251–261. [CrossRef]
13. Saverwyns, S.; Currie, C.; Lamas-Delgado, E. Macro X-ray fluorescence scanning (MA-XRF) as tool in the authentication of
paintings. Microchem. J. 2018, 137, 139–147. [CrossRef]
14. Impallaria, A.; Petrucci, F.; Chiozzi, S.; Evangelisti, F.; Squerzanti, S. A scanner for in situ X-ray radiography of large paintings:
the case of “Paolo and Francesca” by G. Previati. Eur. Phys. J. Plus 2021, 136, 126. [CrossRef]
15. Alberti, R.; Frizzi, T.; Bombelli, L.; Gironda, M.; Aresi, N.; Rosi, F.; Miliani, C.; Tranquilli, G.; Talarico, F.; Cartechini, L. CRONO: A
fast and reconfigurable macro X-ray fluorescence scanner for in-situ investigations of polychrome surfaces. X-ray Spectrom. 2017,
46, 297–302. [CrossRef]
16. Campos, P.H.; Appoloni, C.R.; Rizzutto, M.A.; Leite, A.R.; Assis, R.F.; Santos, H.C.; Silva, T.F.; Rodrigues, C.L.; Tabacniks, M.H.;
Added, N. A low-cost portable system for elemental mapping by XRF aiming in situ analyses. Appl. Radiat. Isot. 2019, 152, 78–85.
[CrossRef]
17. Hocquet, F.P.; Calvo, H.; Xicotencatl, A.C.; Micha, E.; Strivay, D. Elemental 2D imaging of paintings with a mobile EDXRF system.
Anal. Bioanal. Chem. 2011, 399, 3109–3116. [CrossRef]
Sensors 2021, 21, 1913 13 of 13

18. Polese, C.; Dabagov, S.; Esposito, A.; Liedl, A.; Hampai, D.; Bartùli, C.; Ferretti, M. Proposal for a prototype of portable micro-XRF
spectrometer. Nucl. Instrum. Methods Phys. Res. Sect. Beam Interact. Mater. Atoms 2015, 355, 281–284. [CrossRef]
19. Shugar, A.N. Handheld Macro-XRF Scanning: Development of Collimators for Sub-mm Resolution; The Institute of Analytical Philately:
Akron, OH, USA, 2020; Chapter 3, pp. 13–19.
20. Barcellos Lins, S.A.; Ridolfi, S.; Gigante, G.E.; Cesareo, R.; Albini, M.; Riccucci, C.; Carlo, G.; Fabbri, A.; Branchini, P.; Tortora, L.
Differential X-Ray Attenuation in MA-XRF Analysis for a Non-invasive Determination of Gilding Thickness. Front. Chem. 2020,
8, 175. [CrossRef] [PubMed]
21. Barcellos Lins, S.A.; Bremmers, B.; Gigante, G.E. XISMuS—X-ray fluorescence imaging software for multiple samples. SoftwareX
2020, 12, 100621. [CrossRef]
22. Brunetti, A.; Sanchez Del Rio, M.; Golosio, B.; Simionovici, A.; Somogyi, A. A library for X-ray-matter interaction cross sections
for X-ray fluorescence applications. Spectrochim. Acta Part B At. Spectrosc. 2004, 59, 1725–1731. [CrossRef]
23. Guerra, M.F. The Study of the Characterisation and Provenance of Coins and Other Metalwork Using XRF, PIXE and Activation Analysis;
Elsevier: Amsterdam, The Netherlands, 2000; pp. 378–416. [CrossRef]
24. Schiavon, N.; de Palmas, A.; Bulla, C.; Piga, G.; Brunetti, A. An Energy-Dispersive X-Ray Fluorescence Spectrometry and Monte
Carlo simulation study of Iron-Age Nuragic small bronzes (“Navicelle”) from Sardinia, Italy. Spectrochim. Acta Part B At. Spectrosc.
2016, 123, 42–46. [CrossRef]
25. Wang, Z.; Bovik, A. A universal image quality index. IEEE Signal Process. Lett. 2002, 9, 81–84. [CrossRef]
26. Nilsson, J.; Akenine-Möller, T. Understanding SSIM. arXiv 2020, arXiv:2006.13846.
27. Alfeld, M.; Janssens, K.; Dik, J.; De Nolf, W.; Van Der Snickt, G. Optimization of mobile scanning macro-XRF systems for the in
situ investigation of historical paintings. J. Anal. At. Spectrom. 2011, 26, 899–909. [CrossRef]
28. de Waal, D. Micro-Raman and portable Raman spectroscopic investigation of blue pigments in selected Delft plates (17–20th
Century). J. Raman Spectrosc. 2009, 40, 2162–2170. [CrossRef]
29. Vieira Ferreira, L.; Casimiro, T.; Colomban, P. Portuguese tin-glazed earthenware from the 17th century. Part 1: Pigments and
glazes characterization. Spectrochim. Acta Part A Mol. Biomol. Spectrosc. 2013, 104, 437–444. [CrossRef]
30. RRUFF Database. Available online: https://rruff.info/quartz/display=default/R150074 (accessed on 20 Novembre 2020).
31. Pereira, M.; de Lacerda-Arôso, T.; Gomes, M.; Mata, A.; Alves, L.; Colomban, P. Ancient Portuguese Ceramic Wall Tiles
(“Azulejos”): Characterization of the Glaze and Ceramic Pigments. J. Nano Res. 2009, 8, 79–88. [CrossRef]
32. Sakellariou, K.; Miliani, C.; Morresi, A.; Ombelli, M. Spectroscopic investigation of yellow majolica glazes. J. Raman Spectrosc.
2004, 35, 61–67. [CrossRef]
33. Rosi, F.; Manuali, V.; Miliani, C.; Brunetti, B.G.; Sgamellotti, A.; Grygar, T.; Hradil, D. Raman scattering features of lead
pyroantimonate compounds. Part I: XRD and Raman characterization of Pb2 Sb2 O7 doped with tin and zinc. J. Raman Spectrosc.
2009, 40, 107–111. [CrossRef]
34. Sandalinas, C.; Ruiz-Moreno, S.; López-Gil, A.; Miralles, J. Experimental confirmation by Raman spectroscopy of a Pb-Sn-Sb
triple oxide yellow pigment in sixteenth-century Italian pottery. J. Raman Spectrosc. 2006, 37, 1146–1153. [CrossRef]
35. Picolpasso, C. I Tre Libri Dell’ Arte Del Vasajo, Nei Quali Si Tratta Non Solo La Pratica, Ma Brevemente Tutti I Secreti Di Essa Cosa Che
Persino Al Di’ D’oggi E’ Stata Senpre Tenuta Ascosta—Primary Source Edition; Nabu Press: Charleston, SC, USA, 2014; p. 126.