Use of Computerised X-ray Tomography in the Study of the Fabrication Methods and Conservation of Ceramics, Glass and Stone Building Materials

Abstract

This work will review and discuss the use of computerised X-ray tomography (CXT) for analysing ancient, manufactured items, like stone building materials, glass and ceramics. It will consider particular techniques required, and/or of benefit, for CXT of heritage materials, such as special precautions during the experimentation to ensure there is no damage to the materials, special imaging methods such as elemental-specific imaging, and sample-specific image analysis requirements. This study shows how the knowledge of internal features, particularly pores, discerned from CXT can be used to reverse engineer the artefact fabrication process. CXT can be used to obtain information on both the raw materials (such as types and impurities) and fabrication techniques used. These abilities can then be used to establish technological evolution and the incidence of ancient behaviours like recycling and allow the linking of particular items to specific production sites. It will also be seen how CXT can aid the development of effective conservation techniques. This work will also consider how conclusions drawn from CXT data can be amended or augmented by the use of complementary non-destructive characterisation methods, such as gas overcondensation.

1. Introduction

Computerised X-ray tomography (CXT; also called XCT, or simply CT) combines X-ray imaging and tomographical algorithms. While the mathematics needed to reconstruct the images was derived in 1917, the X-ray hardware necessary to physically implement the tomography did not start to come online until the 1970s. It is only relatively more recently that laboratory- and benchtop-sized apparatuses have become more widely available to a broader range of researchers, including materials scientists, archaeometrists, and conservation scientists, for the study of heritage materials.

CXT allows the non-invasive study of the internal structure of otherwise opaque materials [1]. CXT involves allowing X-rays to pass through the sample along many different paths in multiple directions, as shown schematically in Figure 1. For inanimate objects, this is achieved by rotating the sample on a turntable within the X-ray chamber such that it can be viewed from different angles. During the course of the rotation, a series of projections (or ‘shadow pictures’ similar to an old-fashioned medical radiograph) are obtained because the X-ray beam can traverse the sample along multiple trajectories. However, in ‘whole-body scanners’, used to scan humans, the subject, instead, stays stationary, and it is the X-ray source that rotates around the subject. Once the full complement of several 2-D projections has been acquired, the full 3D image reconstruction can be obtained, typically using a filtered back-projection algorithm utilising cone-beam reconstruction [1]. There are also parallel and fan beam modes of operation [2]. This reconstruction produces a stack of 2D cross-sectional images, or slices, for the entire sample, which reflects the best estimation by the algorithm of the internal 3D structure within the sample that would have produced the particular set of 2D projections observed. The individual 2D picture elements are often called ‘pixels’, while the corresponding (typically cubic) volume elements in 3D are called voxels. The 2D slice reconstructions are sometimes referred to as ‘virtual thin sections’ (VTSs) [3]. Since the slicing is conducted virtually using a computer, it can be re-arranged, at whim, along a different set of planes with a chosen normal.

The main advantage of CXT is that it is non-destructive because it does not require the physical serial sectioning of materials required by microscopy methods [4]. It is also an advance over conventional radiography since it provides full 3D reconstructions of quite thick objects rather than just single, complex, superposed 2D projections that need much interpretation. There are some limitations on the objects that can be studied since it is necessary that (at least some of) the X-rays can fully penetrate the object, which precludes thick objects composed of very heavy (high atomic number) metals.

The aim of this work is to provide a review focussed on representative examples of the applications of CXT to particular heritage materials (ceramics, glass and stone building materials) with a view to understanding their fabrication methods and the efficacy of conservation methods. In particular, this work will consider how the ability of CXT to provide a detailed characterisation of the porous structure of such materials permits this, and it will also consider the use of CXT for density and elemental mapping. The structure of this work is such that it first describes more technical details on the CXT method itself, as well as techniques of image analysis. It then considers the applications of CXT to three main types of heritage materials, namely ceramics, glass and building stone, in turn. For ceramics and glass, it describes how CXT data can be used to discover the raw materials used and reverse engineer the fabrication method. This work also considers the abilities of more specialist CXT techniques beyond simple attenuation imaging, including phase contrast methods and K-edge imaging.

2. Methods of CXT

2.1. Image Acquisition

The types of equipment that can be used for CXT range from commercial bench-top apparatus to bespoke apparatus constructed for use with particular samples and synchrotrons. Laboratory micro-CXT equipment can achieve a resolution down to ~100s nanometres. However, synchrotron X-ray tomography can reach resolutions below 50 nm [5], though this is still larger than most mesopores, and the beam energy required may damage the sample. Further, X-ray microscopes equipped with Fresnel zone plate (FZP) optics can reach approximately 10 nm, which is close to the theoretical limit but with a limited field of view [5]. Due to their greater accessibility to many workers, this review will concentrate on the use of bench-top apparatus, though with some mention, too, of the applications of synchrotron CXT.

Image contrast is defined as ‘a measure of the difference [in] optical densities, or grey levels, of neighbo[u]ring regions of an image’ [6]. Obtaining high contrast in an image greatly facilitates the identification and delineation of particular structural features. The contrast obtained depends upon the intrinsic material properties of a sample, such as its chemical composition, density, shape and structure, which affect the level of attenuation of X-rays passing through it. The attenuation of X-rays is due to interactions with the electrons in the atoms within the material and typically follows a Beer–Lambert type law:

$$I=I_{0}exp\left(-\mu x\right),$$

(1)

where I is the transmitted X-ray intensity, I₀ is the incident intensity, μ is an attenuation coefficient related to the electron density of the material, and x is the path-length of the X-rays through the material. Since the X-rays must pass through the sample to be detected, this puts a limit on the path-lengths (and thus size/thicknesses) of particular materials that can be imaged.

The aforementioned attenuation coefficient depends upon the properties of the X-rays and the material they are passing through. The choice of X-ray peak energy also affects contrast since materials have different levels of attenuation depending upon the X-ray photon energy. It is necessary to select an X-ray energy (by tuning current and voltage parameters) that allows most of the beam to penetrate the less dense regions of the sample but some of the beam to be stopped by the higher-density regions rather than have the whole beam penetrate most regions mostly unattenuated or be stopped entirely.

The spatial resolution is a measure of ‘how small an object can be distinguished in [a CXT] image’ [6]. The higher the resolution, the more the image can distinguish between finer-grained objects. The factors that control the resolution obtained include the X-ray tube focal spot size, type and thickness of the intensifying screen (or ‘scintillator’), and the qualities of the photon detector device. Typically, the intensifying screen has the lowest spatial resolution and thus provides the lowest limit possible with the system.

Standard absorption-based imaging can only distinguish constituent particles of different compositions on the basis of differences in the attenuation of X-rays and, thus, the grey level of particular voxels. This can be ambiguous if the electron density of elements within neighbouring phases is very similar. Hence, it can often be difficult to distinguish different tempering agents in ceramics or pigment and opacity agents in glass if they contain similar atomic number elements. However, a different X-ray imaging method, known as K-edge imaging, can remedy this drawback [1]. The X-ray spectra of given individual elements exhibit sharp increases in absorption at a particular X-ray energy, which is just above the binding energy of the K-shell electron. The K-edge energy is characteristic of each element, as will be seen below. If separate CXT images are acquired at X-ray energies from immediately above and below this step in the spectrum, and the difference in intensities between them is obtained by subtracting one image from another, then only voxels with the specific element contained therein would vary significantly in intensity between the two images, and, thus, be made evident in the subtracted image.

CXT images of glasses and ceramics can contain artefacts, and these can sometimes be removed by suitable image processing, as will be discussed in more detail below. Images can show ring artefacts (concentric circles) arising from local detector effects/defects [7] and line artefacts arising from scattering from dense phases and beam hardening [1].

While CXT is applicable to many heritage materials, such as ceramics, glass and stone objects discussed here, there are also some special precautions that should be borne in mind. After very long exposure to X-ray photons, say with very long image acquisition times (to achieve high signal-to-noise), some radiation damage may occur, which can give rise to colour changes in materials, such as glass [4], which also depend on X-ray fluence. Spataro et al. [8] highlighted that the exposure of archaeological ceramics to significant doses of ionising radiation, such as that used in industrial or lab-based CXT units, will also affect subsequent thermoluminescence dating [9].

2.2. Basic Image Analysis

Image analysis methods can ultimately obtain the size, shape and orientation of components of the material fabric, such as inclusions and voids [6]. The image analysis can also identify spatial variations in density and more abrupt joins between the sub-parts making up a composite vessel, such as coils, slabs, and applied features. Computerised analysis of CXT image data can even detect features that are not obvious through visual inspection alone [6]. Computers can sometimes make much finer distinctions than the human eye, such as between subtle changes in the grey level in the image. All of this information can be used to infer a fabrication method.

Once an image has been obtained, it can have artefacts and noise present that obscure the details. Hence, subsequent image processing is a key step in deriving an accurate model of the void space [10,11]. However, it is first noted that, strictly speaking, the 3D reconstruction obtained of the internal structure of the object is just a model, as it is the output of a mathematical algorithm and not a raw, direct observation, even though it can often be treated as such. The accuracy of the model depends upon the quality, number and frequency (angular step) of 2D projections used to generate it.

While several new image processing algorithms have been introduced, including some based upon artificial intelligence, image processing still involves a significant subjective (human) element, and the techniques involved have been reviewed in detail, especially for CXT, by Guibert et al. [11]. The unavoidable, random noise within an image can be eradicated, to some extent, by the application of a filter, such as a random or Gaussian filter [10,11]. Tomographic data can also contain so-called artefacts, which are spurious features in the image generated by such effects as refractions/scattering and beam hardening (for X-rays). Beam hardening artefacts can be suppressed by the use of an appropriate X-ray filter to remove low-energy X-rays during image acquisition, such as copper filters [3]. For porous media known to possess voids, in order to convert a pore-scale continuous, greyscale image into a discrete void space rendering, it is necessary to segment the image into clearly defined solid and void regions. This classification of the image voxels as different phases (solid and void) requires a segmentation algorithm to decide upon the appropriate allocation of each image voxel. However, this procedure often still involves several subjective elements [10,11]. The segmentation algorithm can, for example, require the making of an assumption about the basic geometry of individual pores to aid identification of the pore boundary where the raw image pixel intensities are insufficient alone.

Once obtained, the 3D tomographic reconstructions (stack of 2D slices of voxels) of the internal structure are available for inspection. This can be basic human visual inspection, but the abundance of information in the 3D reconstruction often makes this unfeasible. Hence, computerised analysis can be used to extract information from images that are ‘beyond the capacities of human visual perception’ [6]. It can extract information on image ‘texture’, which is the spatial distribution of grey-level values within the whole image or region-of-interest (ROI), and ‘gradient’, which characterises the directional trend in these grey-level values. The statistical descriptors of the texture include the Fourier-based first moment of the power spectrum (which measures texture fineness or coarseness), the root-mean-square (rms) variation, the degree of randomness in the grey levels (a measure of ‘entropy’), and fractal-based measures. The potential uses of image analysis will be discussed below.

3. Ceramic Materials

3.1. Determination of Fabrication Technique

3.1.1. Features Detected by CXT

CXT is frequently used to determine the fabrication technique used to form a ceramic vessel. The typical steps in pottery manufacture are gathering and preparing the clay, formation/shaping, applying surface treatment, and firing [12]. It is necessary to use 3D data from tomography to see inside a pottery vessel and to see the form of the internal surface and wall structure to determine its fabrication method. This is especially necessary since treatments of the exterior wall, such as smoothing and application of slip, can greatly conceal the basic fabrication method [12]. As described by various workers [3,13], VTSs from CXT can be used for:

• Detection of joins between different basic construction units (CUs), such as coils;
• Detection of joins between different constituent parts of the vessel, known as construction parts (CPs), such as bases or handles;
• Mapping of voids remaining after burning out of organic temper, or the thermal decomposition of inorganic tempers like carbonates;
• Mapping of pores and their orientations within the ceramic fabric itself;
• Detection of non-plastic inclusions.
• CXT can also be used to discover internal features that result from other features of pottery manufacture or later treatment, such as:

  ▪

  cracks that form during various phases of drying;

  ▪

  slips applied;

  ▪

  sediments;

  ▪

  secondary mending operations.

Kozatsas et al. [3] highlighted that the variation of the orientation of the normal of the VTS and comparing the resultant slices that are possible with CXT enables a better understanding of the nature and disposition of CUs, CPs, etc., which is not possible with fixed physical sectioning. CXT also allows the tracing of evidence for the key aspects of the manufacturing process of ceramics, namely:

• the initial form of the CUs;
• the source of energy for firing;
• the power, direction, and type (compressive or shear, discontinuous or continuous) of the pressures exerted;
• the tools used in interventions;
• the humidity of the clay.

3.1.2. Use of Experimental Archaeology

Experimental archaeology provides a way in which interpretations of features within CXT images of ceramics to imply particular fabrication routes can be validated. This is because the different modes of forming a vessel each leave their own distinctive signature in the item [14,15]. During the fabrication of a vessel, the shaping and squeezing of the clay leads to the formation of a large number of pores. The orientation of these pores can inform on the technique that the potter was using. For example, Lindahl and Pikirayi [14] simulated the fabrication of clay vessels using four different forming methods, namely the U- and N-techniques of coiling and two types of modelling techniques. The first modelling technique is simply to shape a vessel by pressing the walls out of a lump of clay with the fingers. The second modelling technique involves initially pulling a thick coil of clay, using the fingers of one hand on the inside of the vessel with the other hand acting as a support on the outside. Lindahl and Pikirayi [14] found that there was a clear difference in the disposition of the pores depending upon whether a coiling or modelling technique was used. A clear difference was also observed between the U- and N-techniques of coiling. While both coiling methods give rise to clearly visible coils and pores orientated from one wall surface to the other, the U-technique gives a distinct curving of the pores, while the N-technique leads to a diagonal orientation. In contrast, it is more difficult to distinguish between the two aforementioned modelling techniques using two-dimensional sections. However, these findings from experimental archaeology have been used to aid in distinguishing the forming technique used between Early and Late Iron Age pottery from southern Africa [14]. In particular, pottery sherds from the former showed a clear coiling technique, while the latter had a pore structure parallel to the vessel walls, thereby implying a modelling technique was used.

Gait et al. [16] also used experimental archaeology to show how the particular patterns of orientations of voids, seen in CXT images, can distinguish between coil-built and percussion-built vessels. The patterns of the orientations of voids and of alignment of features within the CXT data for the ceramics were found and described using two different image analysis methods, namely the Orientation Index and spherical coordinate axes. It was found that the CXT coupled with the image analysis could distinguish the two forming methods in a statistically reliable way. The findings from CXT were also validated using neutron tomography [16]. Neutrons have greater penetration power than X-rays and, in contrast, are also scattered by light nuclei like hydrogen, so they can detect organic materials directly. However, resolutions possible with neutron images tend to be lower than for X-rays.

Even a polished section of a freshly broken sherd may not reveal sufficient characteristic features of the pores to correctly distinguish the exact forming technique, such as between particular modelling methods. However, the fully three-dimensional rendition provided by CXT can provide additional information on pore shape and connectivity that is not evident in two dimensions. This fully 3D data can also be used to ‘reverse engineer’ the fabrication route for vessels based on the identification of boundaries between components and even the hand marks from the potter in the CXT images [17]. CXT is also less destructive as a fresh break is not required to find out the information. The use of the results from experimental archaeology in the interpretation of CXT data will be discussed below.

3.1.3. Exemplars of the Application of CXT to Fabrication Technique Determination

Kozatsas et al. [3] have provided a comprehensive survey of the CXT evidence for various fabrication materials and methods used to manufacture several different Middle Neolithic vessels of which sherd remains were found at the site of Sesklo in Thessaly, Greece, such as the example given in Figure 2. They found that the construction method can vary even across the same vessel, i.e., that the potter changes their methods part way through manufacture. An additional (to those described above) forming method involves successive assembly of pre-formed CUs, such as slabs [3]. This technique should result in a network of joints between the many constituent CUs. However, the use of wet clay CUs and the use of strong forces during assembly can cause the homogenising adhesion of the CUs such that the ‘texture’ of the joint is reduced to a mere apparently ordered line of small, isolated voids, but this feature can still be detected in CXT VTSs. Kozatsas et al. [3] highlighted that sometimes the as-received, overall size of the object under study can hamper the achievement of the requisite image voxel resolution to reveal the key features diagnostic of a particular fabrication route because the solution possible for replaceable industrial materials, namely fragmentation, is precluded for heritage materials.

Takenouchi and Yamahana used CXT to discover the shaping techniques used in Predynastic Egyptian pottery [12]. In particular, they also found some samples that used multiple, rather than just one, forming methods. For example, a black-topped beaker used both coiling and slab building in the flat base slab. The CXT data also provided evidence for the use of a turning device for making some pots by detecting rill marks on the rim.

Stark contrasts between the densities of various regions in a sample revealed by CXT images can suggest a particular fabrication method [6]. For example, variations in the relative densities of inclusions and incidence of specific void areas and lower density areas indicated the characteristic patterns expected for the paddle-and-anvil forming technique used in guan jar fabrication of Neolithic Chinese (Longshan) pottery. A statistical evaluation of the aforementioned features enabled the spatial variation of particular production procedures to be discerned. The previously mentioned image analysis descriptors can also be used to extract information related to the fabrication method. For example, the application of the “Beta fractal” analysis to an image of a sherd of Longshan pottery (see Figure 3) revealed the fine detail in the spatial distribution of the density of inclusions and, in particular, showing that higher-density particles were gathered towards the top of a guan jar during vessel forming, which was not visible to normal visual inspection. Further, the entropy-based analysis algorithm, mentioned in Section 2.2, aided in the sharpening and delineation of paddle impact zones in images. Hence, the image analysis can be used to distinguish between different techniques of paste preparation and vessel forming.

CXT has been widely used to examine the nature and distribution of tempering materials in ceramics. The manufacture of clay vessels often requires the addition of a tempering agent to enable them to survive the large shifts in temperature associated with the firing process [18]. Various types of material can be used as temper, including sand, shell, grit and organic matter like vegetation fibres. If organic matter is used, it is typically burnt out in the firing, leaving behind void spaces within the ceramic walls. Traditional methods, such as petrographic analysis of thin sections using polarised light microscopy, are not always capable of identifying the original organic matter [19]. However, the 3D images available via CXT can be used to determine the characteristic shape and size of the original material and then identify it. A number of studies (discussed below) conducted on a wide variety of archaeological ceramics have shown that this can be achieved and that the technique works well.

CXT has been used to image ceramic sherds from an excavation on St. Catherine’s Island, Georgia, USA, with a resolution of 40 μm [18]. The images were then segmented, based upon their grey levels using an automatic filter, into clay matrix, lithic inclusions, and void phases. Image analysis was used to identify different phases, detect the presence and alignment of voids, and quantify the volume, size, shape, diversity, and distribution of each phase. All of the sherds had evidence of voids that once contained interwoven tempering materials. The image analysis showed that 90% of the potsherds had bimodal distributions for the alignment of voids corresponding to either alignment within five degrees of parallel or perpendicular to the main vessel axis. Some sherds contained both lithic fragments (e.g., quartz grit, shell) and voids. The 3D reconstructions from the CXT images showed that most of the sherds from the walls of vessels had several fine layers within them. This suggested that, instead of being fabricated from a single piece of clay with organic fibre temper running in random directions throughout, the vessel walls were made with multiple thin layers of clay, each with organic temper fibres running in a single direction. These layers were then placed on top of each other such that the directions of the fibres therein criss-crossed each other. However, the images of fragments of bases of vessels showed that the orientation of the temper and meld lines in the clay matrix rotated around the central axis of the vessel, suggesting the bases had been constructed by coiling. This work showed the utility of the 3D reconstructions obtained from CXT data for revealing fabrication processes in different regions of the vessels. However, Sanger et al. [18] did suggest that the CXT data struggled to distinguish between the moulding and slab building methods of fabrication.

Kulkova and Kulkov [19] examined Neolithic ceramic sherds from the sites of Rakushechny Yar, Okhta, and Padolijc in Russia. Analysis of CXT images of sherds from Rakushechny Yar revealed a shape and distribution of voids suggestive of the type of organic vegetation, namely Stuckenia pectinate (Sago pondweed), naturally present in the clay of the local floodplain of the lower Don river, while the voids found in the sherds from Padolijc were characteristic of sedges. The CXT images also allowed visualisation of the pore distribution, which, for sherds from Rakushechny Yar, was characteristic of manufacture via coiling, and the uniform distribution of the sand temper (~10%).

Menne et al. [20] examined pottery sherds from Neolithic megalithic tombs in Emsland, Germany, to characterise temper components and non-plastic inclusions. The CXT was able to detect various temper components and inclusions, namely bone, (imprints of) organic straw and biotite, as shown in Figure 4. In particular, while it was not clear whether features seen in thin micrograph sections were simply cavities or non-plastic inclusions, the 3D CXT showed that voids had the particular spatial extent and shape of straw material. Menne et al. also found that the porosity was higher for older samples, which may reflect a change in production technology to obtain denser pots [20]. However, it was not possible to distinguish different vessel types based upon the porosity values alone; otherwise, this would have allowed the identification of the vessel type based even upon a very small sherd fragment, which was otherwise typologically ambiguous.

CXT has been used to examine a variety of ceramic items, not just pottery vessels. Spataro et al. [8] obtained images of Assyrian, cuneiform, clay tablets with a resolution of 75 μm using CXT. The CXT images revealed variations in the incidence of internal cracks/gaps between layers, the quantity and size of mineral inclusions, and the presence of organic remains. Some tablets lacked any signs of plant temper but did show sandy inclusions, thereby confirming complementary data from thin micrograph sections, but that was only for much smaller sampled volumes than was possible with CXT. In contrast, other tablets clearly showed elongated, planar voids left by organic material that were aligned more-or-less in parallel, thereby suggesting the clay was clearly worked in a particular direction during forming. The 3D reconstructions obtained from the CXT data showed the impression of a knotted cord around which one tablet had been formed. Hence, this work provided insight into the forming techniques for clay tablets.

3.2. Studies of Temper Materials Themselves

CXT imaging of the voids left by organic temper in ceramics has not just been used to provide information on the ancient pottery manufacturing processes but also aid research into the temper material itself, including archaeo-botany studies as the potsherds are effectively traps for the preservation of (the forms of) ancient organic material. CXT images of Neolithic (c. 3700-2900 BCE) pottery sherds from the Khashm el Girba site in Sudan have revealed that the 3D voids left by organic temper could be identified as left behind by domesticated sorghum [21]. The domesticated form was identified from the morphologically distinct voids left by domesticated sorghum spikelets, compared with the wild type. This showed that sorghum had been domesticated much earlier than previous archaeological research had suggested.

3.3. Decoration

CXT can also be used to study the decoration of pottery vessels, such as via painting. For example, CXT has helped identify the nature and origins of a curious, unpainted, circular feature within a painting of two panthers from the Archaic period, the Corinthian alabastron [22]. The CXT revealed that the circular feature was underlain by a plug insert made of the same clay as the remainder of the vessel. This insert may have been the result of an ancient repair performed while the vessel was still in the firm leather-hard stage before the final painting was executed. This repair may have been required by the presence of an unsightly and structural-weakening inclusion in the original clay. CXT has also been used to allow the discrimination of heavy-metal-based pigments in the decoration of bowls from the Chinese Xuande Reign period (1426–1435 CE) [23].

3.4. Uses of Ceramics

Ultimately, CXT can also be used to identify the uses of pottery vessels, such as prehistoric Chinese pottery from Changning [24]. CXT has shown that carbonised residues within an amphora were birch bark tar used to make adhesive materials. This tar may have been used, in turn, to make composite tools found in close proximity to the site.

The lack of the decay of the carbonate and shell inclusions due to thermal treatment in Neolithic ceramic sherds from the aforementioned (in Section 3.1.3) site of Rakushechny Yar in Russia suggested that the original pot had not been subject to the repeated firing actions that might be associated with cooking, and so the vessel was probably used for food storage [19]. In contrast, the lack of such inclusions, but with their remnant voids, in sherds from Okhta suggested that they had been fired to temperatures in excess of 780 °C.

3.5. Validation of Typological Classification

Traditionally, pottery is classified using typological analysis. However, the form alone may not be sufficient to allow confident classification. CXT can be used to determine whether differences in paste composition and the forming process correlate with this typological analysis. For example, finds from excavations in the Karst region of northeastern Italy have led to the definition of a local Neolithic facies, designated the Vlaška Group or Cultura dei vasi a coppa, dating to the second half of the 6th millennium BCE [25]. This culture may also have possible links with the Danilo culture in Dalmatia. The particular formal characteristics of typical vases from this culture are bowls that are basically variants of a hemisphere but with differences in the restriction of the upper part, the depth of the body, and the decoration. However, vessels have also been found with unique decorative motifs that have led to doubts over their classification as part of, or influenced by, this culture. Bernardini et al. have studied 13 pottery samples, with the clearest examples of the vasi a coppa type originating from the Zingari and Tartaruga caves, but also a sample, designated 5880, found in the Ciclami cave, that has a variant typology that has some similarities, but also differences, with the Fiorano culture, and thus the typological attribution and origin is uncertain [25]. The pottery samples were imaged with CXT to a pixel resolution of 11.98 to 39.71 μm, and the images were segmented in order to separate out the lithic inclusions (including both lithic temper material and inadvertent lithic components of the paste) (Li), the pores (Po), and the clay matrix (Cl). While a systematic, objective segmentation based upon the histogram of image grey levels was attempted, some more subjective intervention based upon visual inspection was required. Once the satisfactory segmentation of the image cross-sections was achieved, the ratios of the areas of the various phases (Cl + Po/Li and Cl/Li) were obtained, and these showed key differences between the samples. The clear vasi a coppa samples had a more lithic temper, as shown by Cl + Po/Li and Cl/Li ratios of 4.26–12.20 and 4.23–12.15, respectively, while Sample 5880 had Cl + Po/Li and Cl/Li ratios of 212.26 and 206.88, respectively. The lithic temper in the vasi a coppa samples consisted of angular, euhedral calcite fragments and was added to modify the initially too-plastic raw material, while Sample 5880 only had a few small sub-circular lithic grains, probably of silicate minerals already present in the raw material. The presence of the well-preserved calcite in the vasi a coppa samples suggests a relatively low firing temperature of 700–800 °C. Further, the pores in these samples were observed in the CXT images to run parallel to the pottery walls, suggesting a modelling technique was used for forming them based upon previous experimental archaeology [14]. In contrast, the CXT data showed that Sample 5880 had a different production technology and was, thus, probably imported from elsewhere, most likely from an area of the Fiorano culture where the pottery has a similar fine-grained fabric with numerous grog fragments of silicate minerals, and relatively lacking in calcite. Hence, the structural findings from CXT are consistent with the typological comparisons with the Fiorano style mentioned above.

3.6. Special CXT and Image Data Analysis Techniques

CXT has a number of variants that can be used to extract more information than conventional X-ray attenuation images. Automated image analysis algorithms can be used to process the immense amount of data contained within a 3D data set from the CXT of a pottery vessel. These various algorithms can be used to infer fabrication methods.

As mentioned above in Section 2.1, the contrast in the CXT images arises from differences in X-ray absorbance and, then, ultimately, to differences in electron density and atomic number of the elements present. Sometimes, this simple difference in the absorption of X-rays is sufficient to identify separate components of a ceramic matrix and their origins. However, when this is not possible, other more sophisticated imaging methods are available to do this, as will be described below. However, Kahl and Ramminger [26] did find that the X-ray attenuation coefficient for grog, which is previously fired and ground up clay, was a bit lower than that for the newer clay matrix and sufficient to distinguish the two, and thereby, identify the use of grog. However, given the similarity in attenuation of the two materials, straightforward segmentation of the two phases (e.g., based upon a histogram of voxel intensities) was not reliable, and a more sophisticated segmentation algorithm based on iterative steps of noise reduction by nonlinear anisotropic diffusion methods was found necessary.

The conventional CXT method described in Section 1 uses a single polychromatic X-ray beam, which is one containing a distribution of many different X-ray photon energies. However, dual-energy computed tomography (DECT) uses two alternating or interleaved X-ray spectra (typically one of high and one of low energy) to acquire attenuation maps. It then uses mathematical techniques to calculate a number of parameters, such as effective atomic number, which is the weighted average of the atomic numbers of each element present. This technique has been used to show that the clays used to produce Roman Sigillata pottery from Italy were not always well prepared or thoroughly mixed due to the heterogeneities evident in the density plots of effective atomic number variation [27]. In addition, X-ray microfluorescence can be used to reveal elemental compositions of pottery samples [28], albeit only superficially, in contrast to radiography/tomography, where the information is from the whole volume (see below).

The analysis of CXT images is not only conducted using subjective visual inspection but can also include various quantitative techniques. For example, data extracted from CXT images of 20 Late Copper Age bowls from Deschmann’s pile dwellings in central Slovenia and the Triest Karst have been analysed using principal component analysis (PCA) that considered the percentages of lithic inclusions, area, length, width and inclusion/clay ratio as variables, to determine the fabrication method and test the theory of origin [29]. The samples were imaged with a pixel resolution of ~20–40 μm, and the images were segmented into clay matrix, pore, and lithic inclusion phases in a similar way to previous work [25] described above to obtain the areas of each phase in each image. The CXT data allowed both the visualisation and quantification of the lithic inclusions, clay matrix, pores and dis-junctures generally originating from the firing shrinkage, loss of temper and the forming method. The ratio of the areas of each phase (such as lithic inclusions to clay matrix) was distinctive of a particular type of paste used, and the typical sizes of features could distinguish fine- and coarse-grained pastes. The CXT also allowed the identification of particular mineral phases via their crystal shapes. Utilising the findings from experimental archaeology obtained by Lindahl and Pikirayi [14] to interpret the CXT images, it was possible to use the observation that some samples showed pores with diagonal orientation to conclude that they were produced by the N-coiling technique, whereas another sample had pores parallel to the vessel surface indicating it was produced by a modelling technique. It was also observed in virtual sections from the CXT data that, in many samples, the core of the base was surrounded by a circular concentric layer of paste, suggesting it was formed separately by a different technique and then joined to the walls.

However, the PCA allowed the extraction of even more information from the data [29]. The variables in the analysis included the inclusion/clay ratio and percentages of area of each phase, as well as the maximum length and maximum width dimensions of the features of different phases, which were also used to obtain the distributions of these parameters. The PCA revealed that the various samples could be separated into different groupings based upon their characteristic image-derived parameters. For example, it was possible to largely distinguish vessels from Deschmann’s pile dwellings and the Triest Karst based upon this analysis. The results suggested that the sets of samples were produced locally using different recipes and techniques. Hence, this analysis can be used to propose regions of origin for new samples simply based upon the PCA of their CXT data.

PCA has also been used to analyse the CXT data for Bronze Age pottery from the site of Al-Khutm in Oman [30]. While some ceramic materials can have quite different ratios of lithic inclusions to clay, such as those from the northern Adriatic region mentioned above, there are other materials where similar values of this ratio correspond to very different pastes [30]. For example, the same value of this ratio could be achieved via either a few very large lithic inclusions or via an abundance of very small lithic inclusions. For example, two coarse-grained vessels found at the site both were attributed to the Wadi Suq period and had relatively high inclusion/clay ratio (>0.25), but one had abundant temper material with heterogeneous density and sizes up to medium gravel (from <1 to ~8 mm), while the other had abundant low-density temper with size only up to fine gravel (from <1 to ~3–4 mm). These variations can be distinguished in the CXT data. In addition, these coarse-grained vessels were clustered together in the PCA analysis, which distinguished them from another cluster of medium-grained pastes. Several vessels discovered at the site, from both the Umman-Nar and Wadi Suq periods, had numerous elongated empty areas with sub-circular sections, corresponding to voids once occupied by vegetal temper, observed in their CXT images. These voids were also noted to be generally parallel to the wall surfaces and elongated along the circumference of the vessels. This was interpreted to suggest that the vessels were formed on a wheel. Further, no pores of the types (with the appropriate orientations) generally associated with coiling or modelling techniques were observed. However, it is still conceivable that the vessels were formed by wheel shaping of coil-built rough-outs instead, as this has been seen in finds from other sites in the region. Nonetheless, the CXT data has enabled differences in forming techniques to be discerned.

Other types of statistical analysis of CXT images are possible. Once the relevant phase has been identified in the CXT images, the quantitative analysis of the abundance (particle numbers/frequency) and size distribution of inorganic (e.g., rock) temper materials has shown that deliberate sorting of the temper had taken place in the fabrication of late Mesolithic to Neolith pottery from the Hamburg-Boburg site in Germany [26].

A novel image segmentation algorithm has been used to help identify a previously unrecognised pottery-building method called spiralled patchwork technology (SPT) in Neolithic pottery [13]. This is where roughly circular patches are constructed by spiral coiling long pieces of clay, and then the pot is made up by tiling the circular patches together in a slightly overlapping arrangement. The particular spatial organisation of pores within the vessel sherds was determined with the aid of the Hough transform, which is an image analysis tool for the detection of points distributed along straight lines or parametric shapes in images. Analysis of experimental pottery was used to demonstrate that the conventional coiling method was characterised by parallel line patterns of pores and that the same patterns occur in some archaeological samples, while more circular patterns are associated with the newly identified SPT, and this has also been identified in some pottery sherds, as shown in Figure 5.

4. Glass Materials

4.1. Determination of Fabrication Technique

CXT data can be used to visualise the internal structure of a glass object to detect features that enable the determination of the type of raw materials and the method of manufacture of the glass. These features include the 3D shape, size, and number of bubbles and opacifying crystals. For some objects, such as dragonfly eye beads from the Shenmingpu site in China, the glassy surface makes it difficult to determine the materials of construction unless the bead is damaged and the interior is visible [31]. This type of bead can be made from faience (glazed quartz), frit, glazed pottery, or glass, but this is only apparent from observing the interior of the bead. CXT provides a method to do this in a non-destructive way. Yang et al. performed CXT imaging at a pixel resolution of 13 μm on beads from a tomb at the Shenmingpu site [31]. The virtual slices from the CXT showed that the interior structures of the beads could be divided into two main types, one which had many irregular pores, while the other was denser. Some of the first type of beads consisted of crushed quartz (sand) particles bound within glassy phases (faience), while the second type was glazed pottery (clay ceramic). The CXT data also showed that the former had a higher-density glaze, containing heavy elements, surrounding a less dense core of silica with pores. In contrast, frit-based beads had glaze coatings and cores of similar density. Examination of the virtual sections of the quartz beads revealed more of the fabrication process. The transverse cross-section of one bead revealed a circular hole, which had straight sides in the longitudinal section, suggesting the bead had been supported on a cylindrical support during shaping and firing. Further, these sections also showed that the quartz mantle was round, suggesting quartz particles had been piled up on the cylindrical support to form the core of the bead that was subsequently glazed. The external decoration of the glaze, to make it look eye-like, appeared to stand proud of the glaze and was separated by another layer in the transverse sections. Hence, the CXT data were able to non-destructively determine the raw materials and steps of manufacture of the beads.

CXT can be used to visualise the size, spatial distribution and shape of bubbles within the glass, as these are affected by the processes and mechanical movements undergone by the hot glass during manufacture [4]. The main processes that lead to bubble formation in the first place are, for smaller “seed”-type bubbles (<100 μm), the chemical reactions between glasses of different compositions, while, for larger bubbles, furnace dust particles can act as nuclei or entrapment of air can occur during fabrication [32].

Finds of round eye beads from sites in China have revealed two different types of eye bead, namely those with embedded eyes and those with raised eyes standing proud of the spherical core [33,34]. The CXT data showed that, when eyes are formed by pressing new glass in/onto an existing core, the previously spherical bubbles in the core underneath the site of the embedded eye become compressed to ellipsoidal shapes by the pressure applied to embed the eyes [33,34]. Indeed, in locations between two eyes, the bubbles have even become triangular in shape, reflecting pressure coming from two directions. When the eye material is attached to the exterior of the core, some air gets trapped between them, and this is evident in the large bubbles at the boundary between the two [34]. In a different type of Chinese bead in the form of a spindle, the air bubbles within the core of blue glass are elongated along the direction of the long axis of the bead, suggesting it was formed by stretching [33]. However, in contrast, the bubbles, within the yellow glass composing decorative stripes on the beads, were spherical, suggesting the decoration was applied after the stretching of the underlying core, probably into grooves made into that core [33].

CXT studies of bubbles have also shown that, in order to introduce a spiral decoration into Iron Age Scottish beads, it would have been necessary to keep the bead heated to conduct a winding process to produce the spiral [4]. This would have produced a trail of bubbles in the direction of the spiral (as seen in Figure 6C), and the bubbles themselves may even become elongated in the direction of the spiral, although the latter is sometimes not observed if the glass is kept heated sufficiently long for the bubbles to reacquire spherical shape, as occurred in the Scottish beads.

In contrast, the retained elongation of bubbles in the direction of winding, seen in 3D CXT images (with ~9 μm resolution), has shown that glass beads found in a fourth-century Scythian cemetery in southern Ukraine were formed by winding around a mandrel, and CXT, additionally, allowed the ruling out of the fabrication involving folding/bending of the glass [32]. Further, clay-turned ceramic deposits, forming a coating of the mandrel, and traces of tool use were also seen in the CXT images. The CXT data allowed the reconstruction of the sequence of operations of the bead maker.

Elongations of bubbles have also been observed in other glass samples, such as in Roman glass mosaic tesserae (denoted BEY) found during excavations in Beirut, Lebanon, and dated to the first half of the 1st century CE [35]. The bubbles in these samples were studied using FEG-SEM (Figure 7), gas overcondensation (Figure 8), and CXT (Figure 9). Bubbles of different sizes are a common characteristic of Roman glasses [4]. In the CXT slices, shown in Figure 9, for the (yellow) tessera denoted BEY159-Y, it can be seen that the black bubbles are typically not circular but often ovoid or lenticular in shape, and it is also noted that the long axis of the bubbles runs parallel to the orientation of the white and dark bands. The high atomic number (and, thus, high X-ray absorbance) inclusions and streaks in the (green) tessera denoted BEY159-G and the bright inclusions in BEY159-Y (Figure 9) are both lead antimonate (Pb₂Sb₂O₇) crystals. One or two of the bubbles in BEY159-G also show some much smaller distortions from regular, circular shapes. However, the bubbles in BEY159-G are generally much more circular than those of BEY159-Y because the former contains PbO oxide levels of 4%, and the glass, therefore, has a higher melting temperature, leading to shorter working periods. The banding and elongation of the bubbles in BEY159-Y is indicative of the direction of internal folding and stretching of the still molten, or softened, glass before it is cooled, and the patterns are then frozen in [4,35]. The pouring lines observable in the BEY glass tesserae are due to their production process. Glass is first poured into a rectangular mould to the thickness of the glass tesserae and, in so doing, often produces lines of elongation in the tesserae produced. Strips of the desired width of the tesserae are then cut from the block; individual tesserae are snapped off over a sharp edge.

FEG SEM imaging revealed that the Roman glass tesserae are mesoporous, as seen in Figure 7 [35]. The special gas sorption technique involving overcondensation (shown schematically in Figure 8) means gas desorption, though an indirect method, can then probe pore sizes all the way from micropores (<2 nm) to large macropores [1], like the bubbles in Figure 9. This means it can provide a complementary linking technique between the two imaging modalities of FEG SEM and CXT. As such, in previous work, it was used to show that the mesoporosity had not formed due to leaching processes around the macroporous bubbles, as might have been expected [35].

CXT can reveal the presence of heterogeneities in the glass that are associated with the reuse of waste glass, such as the use as a bead core to wind hot glass around (see Figure 6). The use of waste lead glass as a core for winding melt around it is suggested because if, instead, it had been simply mixed into the new melt and then moulded, the lead would have been more evenly distributed throughout the bead. The reuse of older waste glass hints at the value of the material at the time of use [4].

4.2. Detecting Glass Additives and Inclusions

CXT can be used to analyse inclusions in glasses and determine the processes of their formation and potential intention [36]. Ancient glassmakers obtained coloured glass by adding minerals or metals, often in the form of scraps or powders, as chromophores to the molten glass. This operation was typically performed at the second stage of glass fusion, which is when frit blocks of glass, initially created in primary furnaces, were ground up into chunks and then re-melted in so-called secondary furnaces [36]. Roman glassmakers used their empirical experience to select and mix opacifiers and chromophores, in particular. They could also fine-tune the final appearance of the glass by modifying furnace conditions accordingly. Typical examples of opacifiers and chromophores they used were lead antimonite and stannate added to obtain opaque yellow gradations, copper compounds added to obtain colours from blue to green, cobalt compounds added to obtain deep blues, manganese compounds added to obtain a violet colour, and iron compounds added to obtain light blue (Fe²⁺) or yellow-green (Fe³⁺) [36]. Since the X-ray absorbance depends upon electron density, it can usually distinguish regions of such heavy metal opacifiers and chromophores from the less dense glassy matrix but cannot distinguish the exact ionisation state of the chromophore. Standard CXT can thus measure the dispersion, shape, and size of such additives.

CXT was used to examine two several-millimetre diameter chunks of Roman glass found at Aquileia, Italy, that visual inspection showed had large inclusions [36], as seen in Figure 10.

CXT was able to provide a 3D reconstruction and surface rendering of the inclusions, as shown in Figure 11. The 3D reconstruction of inclusion #1 (in Figure 11a) showed it was a metallic particle with a ~150–700 μm thick layer of patina induced by corrosion occurring only upon the surface exposed to the external environment. The CXT data also showed that the metallic core was not continuous but composed of several metallic fragments. Analysis of the image (see Figure 12) showed that, by volume fraction, the inclusion was composed of 18% corrosion patina, 63% metal fragments, and 19% of a further unidentified material (suggested to be glass). For inclusion #2, a similar analysis was conducted on a hemispherical cap and revealed that there was very little corrosion patina, and the interior appeared to be composed of small dense globules with a jagged structure distributed through a much less dense matrix (see Figure 11d–f).

The imaging data permitted the development of a scenario for the formation of the inclusions based upon a theory of what had happened during the glass manufacturing process. Since it was observed, in the CXT images, that there was glass amidst the metal, as well as metal fragments and separated phases, it was suggested that the formation of the inclusions was correlated with the occurrence of a colloidal system in the glass furnace induced by the dispersion of metal powder or scraps within the melted glass [36]. It was further supposed that, if an excess of metal scraps is added to the molten glass during the colouring process, then the metal may not fully disperse throughout the molten glass but, instead, would precipitate or coalesce into aggregates. This aggregation would potentially capture small amounts of glass within the metal packing.

CXT has also been used to study the distribution of pigments and inclusions, composed of heavy elements, in ancient Thai glasses from ~2100–1200 BP [37]. It was found that X-ray opaque material was randomly dispersed throughout the matrix of the glass material in the form of bright spots that were attributed to locations of high lead concentration, as this was the heaviest element arising in complementary EDS data. CXT was used to show that the Dvaravati glasses may have been imported or the relevant technology transfer occurred along the Silk Road.

4.3. K-Edge Imaging of Specific Elements

The K-edge imaging method, described in Section 2.1, has been used to image heavy metal inclusions within early medieval (7th–8th centuries) glass from Susteren in The Netherlands. Since, as previously mentioned, antimony and tin were often used in opacifiers and chromophores, element-specific imaging was used to detect these heavy metals. The ‘mapping’ of the K-edge was conducted using metal foils, and their spectra are shown in Figure 13, which shows a clear increase in absorbance at characteristic energies.

The inclusions in a pale green bead with a red rim (denoted SUST-2) were imaged with CXT. Figure 14 shows a non-specific image with a wide field of view, showing the overall context of an example of an inclusion and a close-up, composite, false-colour, element-specific image. From the broader, contextual image (Figure 14a), it can be seen that the inclusion is embedded in a lighter glass matrix, which contains dark, isolated, ovoid structures that are probably bubbles. The composite element-specific image of the inclusion itself (Figure 14b) shows that the heavy metals are concentrated in the inclusion, with relatively little in the surrounding glass matrix. There is also some relative spatial segregation of tin and antinomy within the inclusion.

A sample of pale green, medieval glass from a bead, also found at Susteren (denoted SUST-3), was also imaged using CXT and the K-edge imaging method. Figure 14c shows a close-up, composite, false-colour, element-specific image. In this case, the inclusions contained mostly antimony with very little tin. The dark circles are, again, bubbles in the glass. The round shape suggests that the glass had not flowed after the bubbles had formed because this would have stretched the bubbles into more elongated structures.

This case study of samples SUST-2 and -3 shows that the spatial distributions of each of the two different heavy metal (tin and antimony) opacifier and chromophore agents can be individually mapped (although presented here as a composite map). Hence, differences in the particular agents used and their degrees of inter-mingling with the glass matrix can be distinguished using K-edge imaging. The data suggests that the dispersion of the agent is more successful in SUST-3 when compared with SUST-2.

It has been seen that K-edge can provide much more specific and detailed information on the occurrence and distribution of particular elements. It can, thus, show the use of opacifiers and chromophore agents of a particular type/character via their different elemental content and distribution. CXT mapping of particular elements might also aid in the detection of the use of recycled glass materials in the wider bulk melt.

4.4. Corroded Glass and Glass Conservation

As briefly alluded to above, CXT can be used for the non-destructive study of the corrosion of glass and, then, for monitoring conservation and restoration processes [7]. CXT can help with obtaining a correct assessment of the corrosion state of glass objects so that the most suitable conservation measure can be selected to aid their long-term survival in a non-destructive way so no further damage is incurred. CXT can also be used to evaluate the degree of success of a given conservation process, such as cleaning.

Mees et al. [7] used CXT to evaluate restoration procedures for corrosion formed under controlled laboratory conditions for model potassium-rich glasses. Artificially corroded samples were prepared either by treatment with an acid solution or in a climate chamber by subjecting them to variable temperature and humidity. Restoration methods tested included cleaning with a scalpel or laser, or treating with heavy metal-bearing acrylate. The CXT was able to show the development of corrosion layers, as these tended to have lower X-ray attenuation than non-corroded glass due to the creation of a so-called ‘gel-layer’, which has lower network modifier ion content and higher water content than uncorroded glass. The thicker corrosion layers are associated with leaching out of material from the glass structure, leading to decreased electron density and, then, X-ray absorbance. The lower X-ray absorbances of corroded regions are often due to the loss of alkali and alkaline earth metal ions. Mees et al. [7] observed that, in laboratory experiments, the leached-out phase could accumulate as a thin, densified layer surrounding the thicker, depleted mantle corrosion layer. The CXT images showed that the crust can be continuous and can reach relatively high thicknesses (~40 μm). However, it is unlikely such a crust would form on archaeological glass samples as the leached material would be removed from the glass surface in soil environments. The CXT data showed that cleaning the corrosion with a scalpel often left the lower part of the corrosion layer in place. However, CXT also revealed that even laser cleaning often also led to heterogeneous removal of the corrosion layer. The doping of acrylate coatings with heavy metals permitted their distribution and penetration in the glass to be determined.

Mees et al. also observed the patterns of corrosion seen in archaeological glass samples [7]. The most common type of corrosion was a thin continuous layer along one or more sides of the glass fragment, with roughly constant thickness and a straight or undulating, and typically very sharp (in the CXT images), boundary with the rest of the glass matrix. However, sometimes, a more discontinuous pattern of corrosion was observed, as seen in Figure 15. In some samples, a more fine-layered corrosion zone was observed, and sometimes, the corrosion layer had started to become detached from the unaltered mantle region.

Franceschin et al. [38] used synchrotron CXT with phase contrast and a resolution of ~1–2 μm to study the multi-layer patina preserved within a groove in ancient (~1st c. BCE to 1st c. CE) glass from Aquileia, Italy. CXT could provide full 3D reconstructions of the layered structure of the patina, which is not possible with 2D sectioning. The CXT data suggested that the temporal progression of the glass alteration had three main stages, namely: (1) formation of a hydrated layer, (2) re-organisation of the aggregated silica nanoparticles in laminated layers, and (3) patina homogenisation and ageing.

CXT has been used to examine the degree of corrosion in Bronze Age (dated to the late 15th and 14th centuries BCE) glass samples from Tel Brak in Syria, one containing cobalt to produce a blue colour (denoted BRAK 14) and another containing ferrous (iron) to produce a translucent brown colour (denoted BRAK 17) [35]. Examples of reconstructed 2D slice CXT images for these samples are given in Figure 16. The dark ovoid structures are bubbles in the glass. The brightness of the inclusions in the image of BRAK14 indicates a high atomic number and are calcium antimonate (Ca₂Sb₂O₇) crystals. The lower X-ray absorbance surface corrosion-affected layer can clearly be seen extending almost all the way around the exterior of the chip of BRAK14, while it is a bit more localised in the chip of BRAK17. It has been suggested [35] that the extensive network of cracks throughout the whole of the bodies of the chips of the BRAK14 and BRAK17 glasses may be indicative of brittleness resulting from insufficient annealing during fabrication [39]. A combination of large heterogeneities and numerous bubbles is often indicative of low-quality workmanship and low-quality glass [39]. The glass in the adjoining zones surrounding some bubbles and the cracks within the BRAK14 and BRAK17 samples show less X-ray attenuation than the surrounding matrix. As mentioned above, this is probably due to the leaching of elements from the glass surface, especially sodium, probably by water ingress, from the burial environment along the cracks in the glass. The lack of durability this exemplifies was probably partly due to the relatively low levels of the network stabiliser, CaO, in the glasses from Tel Brak (3.9–5.8%), together with several other factors such as variable levels of humidity and temperature in the semi-desert burial environment in northern Syria.

CXT can also be used to study conservation processes for glass materials, with particular use having been made of the edge enhancement that occurs between two areas of the sample with different refractive indices for X-rays. For example, the edge enhancement phase contrast technique for synchrotron-based CXT has been used to study the incorporation of consolidants into corroded glass. In particular CXT was used to study the dynamics of the incorporation of consolidants into the cracks in archaeological glass [40]. CXT has also been used to study treatments aimed at removing dark-coloured, manganese-rich stains found in the alteration layer in ancient glass [41]. The CXT was used to provide full 3D monitoring of the progress of the gradual dissolution of manganese oxide bodies within glass.

5. Stone Building Materials

5.1. CXT Applications to Building Stone

CXT is widely used to characterise the void space of rocks, mainly in order to predict multi-phase fluid flow in applications such as oil and gas production or carbon sequestration [1]. It can thus also be used for the pore structural characterisation of building materials, especially to assess the current state and the level of success of conservation methods.

Calcareous building stones often have high porosity, largely accessible from the exterior, and are thus susceptible to the penetration of (rain)water, often rendered acidic from air-borne pollutants such as sulphur dioxide and nitrogen oxides, which can lead to internal decay of the skeletal solid structure by chemical attack, and loss of mechanical integrity and stability. Similar damage can occur due to freeze–thaw action from penetrating moisture or biological degradation [42]. Renovation and maintenance activities for old buildings to increase their lifetime have led to the growth in the use of water-repellent and consolidant agents [42]. In order to prevent and/or ameliorate damage occurring via chemical attack, various different kinds of, typically organic, products, such as fluorinated rubber or solutions of alkylalkoxysilane oligomers, are injected into the void space of the stone to act as a protective coating (against chemical or moisture attack), a barrier (against fluid penetration), or a consolidant (to strengthen already damaged structures). The internal pore network structure of the building stone, with variations in pore sizes and connectivity, affects various physical processes that can impact the rock. The pore network determines how far the acid rain can penetrate and how well the protective coating can block this from happening. The depth of penetration of water-repellent or consolidant agents into building stones can range from a few hundredths of a millimetre to several centimetres, and this depth affects the long-term durability of the treatment [42]. It is also necessary to ensure that, when water repellants are applied, while they prevent access to incoming water, they also do not lead to moisture being trapped behind them within the body of the stone. The 3D reconstruction of the distribution of consolidant can be used to assess how well it supports stresses within the rock that would, otherwise, further damage deteriorated stone. The void space structure also determines the rates and extent of exfoliation arising from the dissolution and precipitation of salts due to condensation, migration and evaporation of water in/from the pores. Hence, it is necessary to know how a particular building stone will respond to different forms of attack and various conservation activities.

5.2. Issues in Making Comparisons between Pore Size Distributions

The pore size distribution (PSD) is a key descriptor since it particularly affects the rates of moisture transport, mechanical properties, and the penetration of additives. However, the particular PSDs (typically weighted by volume or number) derived from image analysis, or more indirect methods, like mercury porosimetry, gas sorption and thermoporometry, are frequently quite different in form. This has often been simply attributed to various known phenomena such as that mercury porosimetry and gas desorption just probe externally accessible porosity (and, therefore, miss out isolated, disconnected voids), and the pore-shielding, or ‘ink-bottle’, effect biases the PSD towards smaller pore sizes, which are not issues for imaging methods [43]. However, it has recently been highlighted that there is a more fundamental issue underlying any differences in PSDs [44]. In fact, there are two distinct forms that the distribution of frequencies of pores can possess for fully interconnected void spaces, and mercury porosimetry and (CXT) image analysis each represent archetypal examples of the characterisation techniques that produce these two different forms [44]. Image analysis methods to derive PSDs are based upon partitioning of the void space up into individual elements with clear boundaries, leading to a discrete distribution of pore sizes. In contrast, more indirect pore structure characterisation methods, such as mercury porosimetry, gas sorption, and thermoporometry, produce data consisting of a continuous spectrum of pore sizes [1]. The difference between such continuous and discrete PSDs, and why they arise, is shown schematically in Figure 17. Typically, the indirect techniques for deriving a continuous PSD make use of a physical phase transition (e.g., gas sorption) or fluid-fluid displacement process (e.g., mercury porosimetry), such that a meniscus separating two phases migrates within the porous medium progressively as the key control variable (e.g., vapour or hydrostatic pressure) is incremented/decremented [1]. This meniscus delineates the boundary between voids of sizes accessible up to the value of the control variable so far, and the volume within the expanding phase indicates the volume of those pores. Since the meniscus only advances incrementally, as the control variable is typically adjusted in small steps, the PSD thereby derived is a (near) continuous spectrum of sizes, particularly if the void space geometry invaded is, say, of conical form (or an analogue) as shown in Figure 17. In Figure 17, for the conical arms of the void space in the ‘3D pore size map’, there are no obvious partitions that could separate ‘pores’ of different sizes that only consist of thin cross-sectional slices of the cone. In contrast, an image analysis algorithm, as used for segmented CXT reconstructions (like the ‘binary 2D image’), is more likely to identify that same structure as representing just one individual pore, as seen in the ‘circle equivalent’ in Figure 17. Hence, as apparent in Figure 17, the two types of methods give rise to very different forms of PSD that are supposed to represent the exact same void space (‘3D pore’ in Figure 17).

The existence of these two potential, clearly distinguishable types of PSD descriptors begs the question of which of the two is the ‘correct’ PSD to use. However, in reality, both forms are only derivative abstractions of the whole, real void space. Although, one of the two alternatives might be preferable for, say, distinguishing between porous materials with a particular character (such as discrete PSDs for spongy rocks), and/or for making predictions concerning particular physical processes occurring within the porous material (such as continuous PSDs for considering capillary ingress of conservation additives).

For example, the pores in building stones tend to be larger macropores, which means CXT is often used alongside complementary techniques such as mercury porosimetry. The comparison of the data from these two methods reveals the various issues with each. Brunello et al. [45] compared PSDs from mercury porosimetry and CXT with a voxel resolution of 10 μm for ‘mock-up’ mortars for cultural heritage. The pore size obtained from the image analysis of the CXT data was calculated as “the diameter of a sphere with the same number of voxels as the object”, similar to that shown schematically (in 2D) in Figure 17. In a comparison of cumulative, volume-weighted PSDs, they found big discrepancies in total specific pore volume and the typical size values of the PSDs between the two. Brunello et al. [45] attributed these discrepancies to the pore-shielding effect and mercury porosimetry missing large pores on the surface of the samples. However, the generally steeper, narrower form of the CXT cumulative PSDs, compared to those obtained from mercury porosimetry, could suggest that the effects shown in Figure 17 could also be affecting the results for the mortars.

5.3. Conservation of Building Stones

CXT has been used to study a variety of building stones from several different locations. Bugani et al. [46] obtained CXT images, at a resolution of 2.6 μm, of (~1–10 mm) samples of a biocalcarenite (Lecce stone) building stone in order to characterise the porous structure and assess the impact of conservation processes. Samples were scanned either in a fresh state or after treatment with one of two polymeric conservation products, namely Paraloid B72 (PB72) (poly(ethyl methacrylate-co-methyl acrylate (70/30))) or Fluoroelastomer (poly(hexafluoropropene-co-vinylidenefluoride)). The images were used to map porosity and its variation due to treatment and/or ageing, as well as the wall thickness of the stone skeleton. It was seen that treatment with PB72 led to a decrease in porosity and wall thickness of thinner pores but an increase in the thickness of already thicker ones. However, only small changes were observed in the PSD between before and after treatment, suggesting there had only been relatively small amounts of polymer ingress. Unfortunately, the CXT spatial resolution was too large to be able to explicitly see where the polymer treatment had gone, which meant it had likely formed thin films or filled pores of sizes below the resolution limit. CXT resolution, with synchrotron radiation, can go down to the tens of nanometres, so explicit detection of the polymer would, in principle, be possible [1]. Further, sometimes, the X-ray absorbance-based contrast between the additive and the building stone is insufficient to distinguish them well in CXT images. In that case, some sort of X-ray contrast agent can be mixed with the additive to enable its position to be detected in CXT images [42]. It is possible to validate that the presence of the contrast agent does not affect the (often transport) properties of the additive being followed by the CXT.

Very high-resolution CXT, with a resolution of ~30 nm, has been used to study Al-tobermorite particles extracted from an ancient Roman seawater harbour concrete sample that was ~9 μm thick [47]. The high-resolution CXT for this study was obtained by putting the sample in a vacuum environment. While the sample itself is not destroyed by the imaging, the sample size required to achieve very high resolution means the sample itself may need to be fragmented.

5.4. Temporally Evolving Systems

Besides just characterising a fixed void space structure, CXT can also be used to study either evolving pore structures, which are changing over time, or dynamic processes that occur within the void space over real time. CXT can be used to follow dynamic processes, such as the ingress of fluids into porous systems if the image acquisition time is sufficiently short compared with the characteristic time of the physical process being monitored. For example, CXT can be used to follow the relatively slow evolution in the pore structures of natural samples of building stones [48] or even following accelerated laboratory-based treatments used to mimic the longer-term, natural exposure to atmospheric acidic pollutants [49,50]. The CXT data can provide valuable information on the modifications to the pore structure induced via chemical attack, such as the formation of interior voids, otherwise hidden via the creation of exterior crusts [49]. CXT can also be used to follow the development of cracks, the locations of crystallisation out of salts, and the particular migration paths of permeating liquids [51]. The detailed surface renderings of objects provided by the full 3D reconstruction from CXT data can be used to follow salt deposition processes in fine detail [50]. Indeed, the spatially-resolved data from CXT can be used to follow the simultaneous combination of deposition within and erosion of the original stone as it occurs in different locations within the same sample, as seen in Figure 18.

The characterisation of rock void spaces in order to run simulations of physicochemical processes therein has been reviewed extensively elsewhere [52]. Much of this work could readily be applied to problems of interest in the conservation of building stones. The 3D structural representation model of the void space generated from the 3D CXT data set can be used as a basis for the simulation of more complex physical processes within the void space that cannot be directly monitored experimentally or to further understand the physical processes involved [52]. These simulation predictions might be validated via complementary techniques that provide information not readily accessible using CXT, such as water uptake into building stones followed by neutron radiography, because the contrast between water and the rock is not sufficient to use CXT itself for this task [53].

6. Discussion

It has been seen that CXT enables the non-destructive study of the interior of irreplaceable heritage materials, in contrast to mercury porosimetry, which can leave behind entrapped mercury, and microscopy, which can require serial sectioning. CXT also allows the full 3D mapping of internal boundaries within solid phases, which is not possible at all with indirect pore-only characterisation methods and is only possible in 2D with microscopy utilising serial sectioning. The 3D mapping capability of CXT especially aids the discernment of the specific traces of particular fabrication methods. This also means that CXT is better at identifying particular tempering agents than light or electron microscopy. Indeed, CXT is much better at identifying specific inclusions than electron microscopy. However, CXT images can contain various artefacts, and so do SEM images, although it is possible to use image analysis algorithms to clean up some of these artefacts in both.

A particular fabrication technique often leads to the introduction or transformation of certain types of pores within a material, the presence of which can then be used to deduce the past use of that technique. It is much harder to determine these critical pore shapes with more indirect methods, compared with CXT. However, the limitations on spatial sampling, via the allowed field of view and sizes of pores detectable due to resolution limits, can mean that CXT will miss some pores. Conversely, it should be noted that CT scanners with relatively low spatial resolution can study very large sample volumes (like human bodies) much bigger than is possible with any of electron microscopy, gas sorption or mercury porosimetry. It has been seen above that the aforementioned issue with CXT at very small length scales (~a few nanometres) can be remedied by coupling CXT with another imaging modality, such as FEG SEM, that has a higher resolution, but integrating the two data sets can be problematic. In contrast, mercury porosimetry and gas overcondensation can both probe pore sizes across a very wide range of length scales (from 100 s microns down to a few nanometres for mercury porosimetry, or molecular scales for gas overcondensation) all in one, simple experiment [1]. This means these complementary methods can provide a linkage between pores detected by different imaging modalities, including some limited information on connectivities and relative spatial distribution of the two sets of pores from each imaging modality [1,35]. However, it should be kept in mind, when selecting the appropriate characterisation technique or comparing data sets, that image analysis methods and indirect pore structural characterisation techniques involving ingress of probe fluids provide fundamentally different types of pore size distribution that may not coincide for some materials with certain types of pore network structures.

CXT can often better distinguish chemical heterogeneity within materials than indirect methods and additionally provide 3D spatial information using special imaging techniques, such as K-edge imaging, without the complex experiments and extensive modelling required with indirect methods. CXT can also study systems evolving over time because the sample does not have to be fixed in some way prior to being imaged, as is often necessary with other techniques.

7. Conclusions

The capabilities of CXT for characterising heritage materials have been surveyed. In particular, it has been seen that CXT has many advantages over traditional methods, such as serial sectioning for microscopy. It has been seen that CXT can be used to determine the raw materials and fabrication methods of archaeological samples of pottery and glass. CXT can provide fully 3D-spatially-resolved information on the distributions of porosity, pore size, and elemental composition, which can be used to deduce many details of the manufacturing route. Some issues with the use of CXT with heritage materials have been highlighted, such as the possibility of sample damage by the X-ray beam. The power of image analysis methods to deliver information not immediately obvious to the operator has been seen. CXT can not only structurally characterise the void spaces of building materials, it can also follow dynamic processes within them in real time. Hence, the implementation and performance of conservation treatments can be followed and assessed in great detail. It has also been seen that CXT can interface in a synergistic way with a variety of complementary methods, such as gas overcondensation or mercury porosimetry, to obtain a more comprehensive characterisation of heritage material, provided the inherent differences in the type of information obtained by each is well understood.