Development of a Macro X-ray Fluorescence (MA-XRF) Scanner System for In Situ Analysis of Paintings That Operates in a Static or Dynamic Method

Abstract

This work presents the development of a macro X-ray fluorescence (MA-XRF) scanner system for in situ analysis of paintings. The instrument was developed to operate using continuous acquisitions, where the module with the X-ray tube and detector moves at a constant speed, dynamically collecting spectra for each pixel of the artwork. Another possible configuration for the instrument is static acquisitions, where the module with the X-ray tube and detector remains stationary to acquire spectra for each pixel. The work also includes the analytical characterization of the system, which incorporates a 1.00 mm collimator that allows for a resolution of 1.76 mm. Additionally, the study presents the results of the analysis of two Brazilian paintings using this instrument. The elemental maps obtained enabled the characterization of the pigments used in the creation of the artworks and materials used in restoration processes.

1. Introduction

Imaging spectroscopic techniques, such as elemental mapping by X-ray fluorescence (XRF), were extensively employed in the investigation of various materials [1,2,3,4,5,6,7,8,9]. Currently, there are commercially available portable and fixed instruments that enable the execution of elemental mapping by XRF experiments, with beam resolutions on the order of micrometers [10,11,12]. The images produced in these experiments have simpler interpretations compared to conventional XRF spectra. As a result, elemental mapping found applications in various fields of knowledge, such as arts, botany, forensics, food, geology, and more.

The use of macro XRF scanning in large areas (MA-XRF) gained significant prominence in the investigation of artworks [13,14,15,16,17,18]. The images obtained through this technique provide valuable information, including restoration regions and underlying paintings, which are highly relevant to conservators and restorers [19,20,21,22,23]. Currently, there are commercially available MA-XRF instruments specifically designed for the investigation of paintings [11,24]. Moreover, various research groups are dedicated to developing portable MA-XRF systems [25,26,27,28,29,30,31]. The development of this technology and the availability of an open system offer the advantage of exploring different experimental conditions throughout the experiments, such as changing the excitation source, which can optimize the detection of specific ranges of elements.

Although MA-XRF is a variant of X-ray fluorescence (XRF), it is a well-established technique that utilizes accessible and well-known experimental instrumentation [32,33,34,35]. However, developing MA-XRF systems still presents challenges. In this case, it is necessary to integrate the module containing the X-ray tube and detector into a cartesian movement system controlled by motors. Achieving this movement with a step size comparable to the beam size, which can be in the order of hundreds of micrometers in some cases, it is one of the challenges. Another challenge involves automating the data acquisition process from the X-ray detector, which must be synchronized with the scanner’s movement. Additionally, the collected data must be organized in a suitable structure for processing in software tools such as PyMca, which enable MA-XRF image reconstructions. Both stages require a comprehensive understanding of automation and programming. Therefore, developing an MA-XRF system necessitates a multidisciplinary team of physicists, engineers, programmers, and other experts.

This study presents the development of a portable MA-XRF system. The instrument incorporates a module with the X-ray tube and detector, capable of operating in static or dynamic modes. In static mode, the instrument pauses at each pixel to collect data, while in dynamic mode, the module continuously moves for data acquisition. The system was specifically designed for in situ analysis of artworks and enables scanning an area of 900 mm × 900 mm with 500 µm steps. In addition to characterizing the developed system, the study shows the results of in situ analyses in Brazilian paintings using this instrument. One painting was analyzed in static mode, while the other was analyzed in dynamic mode.

2. Materials and Methods

Module with X-ray Generator and Detector

To mount the module containing the X-ray source and detector, as shown in Figure 1, an X-ray generator of the MAGNUM cable model from Moxtek was utilized [36]. The generator, equipped with a Cu target, allows for operation at a maximum voltage and current of 50 kV and 200 µA, respectively. These parameters can be adjusted using an external controller. To collimate the X-ray beam, a conical-shaped collimator made of brass, with a 1,60 mm exit hole was attached to the tube. Additionally, to further reduce the beam size for sample irradiation, a brass pinhole with a diameter of 1.00 mm was manufactured and affixed to the collimator’s output.

The detector employed was the 123 FAST SDD model from Amptek, featuring an active area of 70 mm² and a resolution of 122 eV for 5.9 keV energy. The detector has a Be window of 12.5 µm thick and can handle a count rate of up to 10⁶ cps [37]. The detector was externally collimated using a piece with a 17 mm² opening. To access the detector’s functionalities and perform automation processes such as spectral multi-acquisitions, the C# language was utilized. The detector’s cooling is carried out by a system based on the Peltier effect, also known as thermoelectric cooling. This system uses Peltier modules, which create a temperature difference by transferring heat between two sides of the module when an electric current is applied. This approach allows for precise and efficient cooling of the detector, keeping it within an ideal temperature range to ensure stability and measurement accuracy.

The mechanical project of the movement system, where the module with the X-ray tube and detector is integrated, can be seen in Figure 2. This in-house-developed cartesian movement system features a structure made of aluminum, and the movement is performed by Leadshine servo motors. Repeatability tests demonstrated that the system enables 500 µm steps to be executed with an accuracy of 20 µm. The motor used has a nominal torque of 4 N·m, providing sufficient strength for applications that require consistent and reliable performance. It is coupled with a high-resolution encoder, which ensures exceptional precision in movements, allowing for fine and exact control of position and speed. This combination guarantees rapid responses and precise positioning, making it ideal for automation systems and industrial applications that demand high accuracy.

In addition to conducting tests to characterize the system’s resolution, detection limit, and stability, the instrument was utilized for in situ analyses at the Victor Meirelles Museum in Santa Catarina and the School of Fine Arts in Rio de Janeiro, both located in Brazil.

The XRF spectra recorded for each pixel of the analyses were stored in an HDF structure, and the data cube was analyzed using PyMca and Datamuncher software Py/mca version 5.9.3 and datamuncher version 1.6 [38,39]. The Datamuncher software was used to reconstruct the images and perform stitching processes, especially in cases where measurements were taken in different quadrants of a painting. This approach is essential for integrating the captured parts and creating a cohesive image, ensuring that the transitions between the quadrants are smooth and precise. The data storage script was implemented in C#, which also involved the development of tools to control the data movement and acquisition system.

The images of the elemental maps are obtained from the peak area of the respective elements detected in the XRF spectra at each pixel. The contrast in the images is directly related to the value of these areas, with the element having the largest area presented in a gray tone on the grayscale, while the smallest area will show a region in black tones. To obtain the areas, a fitting model was initially constructed from the resultant spectrum obtained by summing all XRF spectra for each pixel. The fitting model was created using PyMca software, covering the region between 2 keV and 16 keV, and includes all the elements detected in the analyses. Beyond 16 keV, only the contributions of scattering peaks (Compton and Rayleigh) from the X-ray tube and sum peaks were detected. The areas of the selected elements were calculated using Datamuncher software, integrating the spectrum at each pixel and applying the pseudo-Voigt function (a linear combination of Gaussian and Lorentzian curves) [40]. An analytical function for the continuum was not selected, allowing for estimation directly from the spectral data, while the background was computed using the sensitive non-linear iterative peak clipping (SNIP) method [41,42,43].

In Figure 3, the fitting model constructed from the scanning data of the painting “A Morta” (Detail please see Supplementarty Materails) from the collection at the Victor Meirelles Museum in Santa Catarina, Brazil, is presented. To minimize errors related to the calculated net areas, special care was taken during the execution of the experiment. For instance, the distance between the painting surface and the instrument was kept constant during the scanning process. The variation in the detector’s dead time was also continuously monitored; this was carried out to verify fluctuations in the obtained counts, which could lead to differences in the net areas of the peaks. Additionally, during the construction of the fitting model, parameters such as detector, beam, peaks, peak shape, attenuators, and matrix were carefully selected in the PyMca routines. These parameters were chosen to closely match the actual experimental conditions, based on the instrument’s technical specifications and after testing. The parameters were adjusted to achieve the smallest χ² values, minimizing the error in the calculated areas.

3. Results and Discussion

3.1. Analytical Characterization of the System: Resolution, Stability, and Detection Limit

The module containing the X-ray tube and detector was positioned, as seen in Figure 1, forming a convergence angle of 60° between the centers of the components. The distance from the center of the pinhole attached to the tube to the projected point of convergence was chosen to be 12 mm. This distance ensures safe operation during the scanning process, preventing any contact with the artifact being analyzed. To verify this operating distance, an ultrasonic position sensor was coupled to the module with the X-ray tube and detector.

To assess the spatial resolution of the scanner, the knife edge method was employed using data obtained during the in-line scanning of a thin copper sheet [44]. Tests were conducted at distances of 12 mm and 13 mm, with the results shown in Figure 4. In this analysis, the profile derived from the Cu-Kα counts graph was fitted with a Gaussian curve, and the resolution was determined as the full width at half maximum (FWHM) of the Gaussian curve. As seen in the graphs in Figure 4, the FWHM values for distances of 12 mm (Figure 4A) and 13 mm (Figure 4C) were 1.76 (Figure 4B) and 1.83 (Figure 4D), respectively. These spatial resolution values are consistent with what is reported in the literature for MA-XRF systems utilizing pinholes [24].

Stability, sensitivity (S), and limit of detection (DL) tests were conducted using a standard steel sample with the following mass fractions: 0.6744 Fe, 0.2242 Cr, 0.0550 Ni, 0.0326 Mo, 0.0106 Mn, 0.0012 V, 0.0020 Cu, 0.0002 Nb, and 0.0004 W. The sample has a density of 7.76 g/cm³ and a thickness of 3.32 cm.

In the stability tests, spectra were collected from the standard sample at intervals of 1 s for 30 min, resulting in a total of 18,000 spectra of sample at the same point. During these tests, the X-ray tube operated at a voltage of 40 kV and a current of 200 µA.

In the stability test, oscillations in the Fe-Kα photon counts were investigated, as seen in the graph in Figure 5. The counts obtained ranged from 200 to 260, with the highest density of counts being around 240 (Figure 5A). These results indicate stability in the XRF spectra (Figure 5B) detection process over the operation time. The majority of the spectra exhibited similar counts, and the range of variation between the maximum and minimum values collected was 23%.

This test holds significant importance, particularly as the scanning process can take hours. Ensuring stability in the data acquisition process is crucial to avoid variations in the contrast of elemental maps.

$$S_i={\displaystyle \frac{N_i}{C_i\times t}}$$

(1)

$${DL}_i=3\times {\displaystyle \frac{\sqrt{N_{back}}}{\sqrt{N_i}}}\times C_i\times \sqrt{t}$$

(2)

The sensitivity ($S_i$) and detection limit (${DL}_i$) were determined using Equations (1) and (2), where N_i and N_back represent the net areas and background intensity of element i, respectively. These values were obtained by fitting models developed in the PyMca software [39]. $C_i$ denotes the concentration of element i, and t represents the acquisition time of the spectra, which were collected over a period of 10 s.

The sensitivity results shown in Figure 5C indicate that for elements with higher concentrations, such as Fe (Z = 26) and Cr (Z = 24), the sensitivity was measured at 68 cps/(mg/g) and 77 cps/(mg/g), respectively. Minor elements, such as V (Z = 23) and Cu (Z = 29), exhibited sensitivities of 34 cps/(mg/g) and 182 cps/(mg/g), respectively. The limit of detection for elements with atomic numbers ranging between 21 < Z < 42 was found to be between 1 and 9 ppm (Figure 5D). The sensitivity and detection limit of the instrument are similar to the results of works reported in the literature that developed MA-XRF portable systems [11,24,28,29].

3.2. Paintings Analysis

The system was used to investigate two paintings: “São Paulo” from the collection of the D. João VI Museum at the School of Fine Art, Federal University of Rio de Janeiro, and “A Morta” from the collection at the Victor Meirelles Museum in Santa Catarina, Brazil. The painting “São Paulo” (Figure 6) dates to the 16th century and its authorship is unknown. Ever “A Morta” (Figure 7) is from the 19th century and is attributed to the Brazilian painter Victor Meirelles.

The painting “São Paulo” was investigated in dynamic mode where the system moved continuously with a speed of 21 mm/s during spectral acquisitions. In this case, for every 1 mm of the artwork, a spectrum was acquired for 40 ms. In this painting, an area of 1122 mm × 288 mm was scanned, totaling a measurement time of approximately 4 h. In the artwork “A Morta”, an area of 250 mm × 220 mm was scanned in static mode, where every 1 mm of step, the system stopped and acquired a spectrum for 50 ms. With this arrangement, the measurement time for the painting “A Morta” was approximately 2 h. Both measurements were performed with the tube operating at a voltage of 40 kV and a current of 200 µA.

Figure 8 shows the spectra with the highest intensity collected from the two analyzed paintings. As expected, the XRF spectrum intensity obtained in static mode is higher than that in dynamic mode. This can be attributed to the longer photon collection time in the automation algorithm, as well as the exclusion of data processing time in the detector, which occurs during the two-point steps.

In dynamic mode, taking a step of 1 mm at a speed of 21 mm/s requires 50 ms. Considering a detector processing time of approximately 10 ms and a spectral acquisition time of 40 ms, the total time amounts to 50 ms. It is worth noting that this difference in intensity is also influenced by the matrix effect in X-ray fluorescence, which is associated with the composition of the analyzed sample. The painting “São Paulo” is on a wooden support, while “A Morta” is on canvas, thus contributing to variations in intensity of spectra due to the support material and the overall composition of the paintings.

A fitting model was developed based on the sum spectra collected from all pixels of the painting “São Paulo”. This model enabled the generation of Pb-L, Cu-K, Hg-L, Ca-K, Fe-K, and Cr-K elemental maps. The maps shown in Figure 9 indicate the presence of lead pigment throughout the artwork, particularly in the white background, suggesting the use of lead white pigment [2PbCO₃·Pb(OH)₂]. This is to highlight that lead pigments are commonly employed as ground for polychrome applications [45].

On the other hand, the Cu-K elemental map associates copper-based pigments with regions of blue and green tones, indicating the presence of azurite [Cu₃(CO₃)₂(OH)₂] and malachite [Cu₂(CO₃)(OH)₂] pigments [46]. The Hg-L elemental map corresponds to flesh regions, where vermilion [HgS] pigment is commonly used [22].

Additionally, the Ca-K, Fe-K, and Cr-K maps were obtained, and their image patterns suggest the application of pigments containing these elements as interventions. Calcium is seen as isolated points in different regions, potentially indicating the use of gypsum (CaSO₄) to fill areas of paint loss. It appears iron was used for interventions in the red (book), blue (mantle), and green (mantle) regions. The chromium image indicates its presence in the green areas, likely associated with chromium oxide (Cr₂O₃) and/or Viridian (Cr₂O₃·2H₂O) pigments.

In the painting “A Morta”, the fitting model for image reconstruction was applied using the sum spectra from all pixels. The resulting elemental maps include Pb-L, Ca-K, Fe-K, and Ti-K (Figure 10). The lead and calcium maps show a widespread presence throughout the painting, allowing us to associate lead pigment with the ground layer and the pigments used in the polychromy, such as inn face region. Calcium, on the other hand, can be attributed to its use as a whitening agent, possibly in the form of calcite (CaCO₃) mixed with other pigments.

The Fe-K image appears in the brown and black regions, indicating the use of Siena (α-FeOOH) and magnetite (Fe₃O₄) pigments, respectively. Notably, previous studies investigating Brazilian artworks from the 19th century, including the artworks of painter Victor Meirelles, found similar pigments [47,48]. These studies also indicate that Brazilian painters during the 19th century had a limited palette, primarily utilizing iron-based pigments in their compositions [23].

The Ti-K elemental map corresponds to the regions where lead is absent, suggesting the use of titanium-based pigments as interventions.

In the stability tests of the X-ray tube and detector configuration (Figure 5A), a 23% difference between the maximum and minimum counts was observed. However, in the elemental maps presented throughout the study, no significant contrast variations were observed in the images, which would suggest that this difference in counts from the X-ray tube and detector stability tests did not affect the instrument’s recorded results. The elemental maps confirm the stability test results, showing a photon Fe-Kα count density of around 240, with the variation between the maximum and minimum values occurring at a low frequency. Therefore, we consider that the observed variation does not compromise the quality of the data obtained.

It is important to highlight that the system is configurable to operate in both dynamic and static modes. The painting “A Morta” was analyzed using both modes. As expected, the results from the dynamic mode produce an image with lower pixel intensity than the static mode illustrated in Figure 10. However, the maps obtained in both modes displayed comparable levels of detail.

4. Conclusions

The results of the analytical characterization and tests on paintings presented demonstrate the effectiveness of the MA-XRF system developed in this work for the investigation of artworks. As shown, the system allows achieving a lateral resolution of 1.76 mm; however, this resolution can be adjusted by using pinholes of varying dimensions. Another option for enhancing the resolution is to incorporate a polycapillary optic into the tube, which is a planned upgrade for the system. This enhancement will enable the scanning of small paintings with greater detail and higher resolution.

Another planned upgrade for the system is the addition of an extra detector to the module, which will decrease the dwell time per pixel and, consequently, reduce the total scanning time.

In addition to the hardware advancements, notable computational work was undertaken. System control tools were developed, enabling any user, after being trained in software operation, to utilize the system for painting analysis.

The MA-XRF system developed in this project was designed to be portable. Despite its large scanning area, the support base of the movement system can be disassembled and reassembled with relative ease. This is due to the entire structure being constructed using aluminum structural profiles, which are easy to handle. It is worth mentioning that the equipment’s movement system can be attached to, for instance, a scaffolding, enabling the analysis of paintings located at great heights.

Regarding the investigation of the paintings, the elemental maps demonstrate the effectiveness of the equipment, as they allow for the identification of the materials used in the creation of the artworks, as well as those resulting from any intervention processes. These images provide valuable insights into the composition of the paintings and the various elements involved in their production and subsequent modifications.