Abstract

The application of state-of-the-art detector systems at the Advanced Neutron Tomography and Radiography Experimental System (ANTARES; Munich, Germany) has led to significant improvements in spatial resolution and contrast for geomaterial imaging. Resolutions of ∼16–100 μm are now possible with fields of view of 33–205 mm, a level now comparable with X-ray computed tomography (XCT), for which a micro-XCT at the Institute for Mineralogy, Crystallography and Material Science at the University of Leipzig was used. Fine pixel resolution comes at the cost of image quality and increased exposure time, so that the optimum configuration for each sample must be determined on a case by case basis. Our interdisciplinary approach has yielded an efficient system of data acquisition, processing, and quantification that is well suited for geomaterial imaging. We expect to find applications in a much wider spectrum of geomaterial research, including the formation of natural glasses, the characterization of limited and/or precious samples such as scientific drill cores, and biomineralization studies.

INTRODUCTION

Neutron computed tomography (NCT) is a nondestructive three-dimensional (3D) imaging method. A collimated thermal neutron beam generated by fission of 235U in a nuclear reactor passes through the specimen. The neutrons can be coherently or incoherently scattered, or absorbed by atomic nuclei within the sample. The remaining neutron flux is imaged using a scintillation screen placed behind the sample. Nuclear reactions between the neutrons and the scintillator material generate photons that are detected by a charged coupled device (CCD) camera (Fig. 1). The sample is on top of a manipulator, which is rotated by a fraction of a degree between each exposure. The resulting set of radiographs is then reconstructed using the inverse radon transformation (Deans, 1983) to generate a 3D image of the local attenuation within the object (Banhart, 2008).

The mass attenuation coefficient may vary by several orders of magnitude without any obvious regularity with atomic number, in contrast with what is observed with X-rays (Fig. 2). Hydrogen highly attenuates neutrons, while it is almost invisible for X-rays; the opposite is true for aluminum. NCT is particularly well suited for imaging hydrous minerals (and glasses) as well as hydrogen-bearing fluids (Carlson, 2006). In this way, the two methods provide complementary information about the sample.

Until recently (Frei, 2009), state-of-the-art NCT imaging could only achieve a spatial resolution of 200 μm, which is too coarse to image many of the fine structures of interest in the geosciences, e.g., crystals, cracks, and pores. In recent years, a number of qualitative feasibility studies have been carried out to test NCT as a method for the geosciences and related fields (Wilding et al., 2005; de Beer et al., 2005, 2006, 2004a, 2004b; Winkler, 2006; Winkler et al., 2002; Fiori et al., 2006; Kardjilov et al., 2007; Sutton, 2008; Bishop et al., 2006; Long, 2007; Nordlund et al., 2001; El Abd et al., 2009; Fijal-Kirejczyk et al., 2009; Milczarek et al., 2005; Zoladek et al., 2008; Hameed et al., 2009; Lehmann et al., 2004; Monteiro et al., 2009; Helfen et al., 2005). Some studies included a comparison with X-ray computed tomography (XCT), which at the time outperformed NCT due to the relatively poor neutron beam collimation and detector technology (Masschaele et al., 2004; Schwarz, 2005; Lehmann et al., 2005).

In this paper we detail recent hardware developments at the Advanced Neutron Tomography and Radiography Experimental System (ANTARES; Munich, Germany) beamline that have turned NCT into a significantly improved tool for the Earth sciences. The application of state-of-the-art scintillator screens now allows us to achieve spatial resolutions as low as 16 μm. Our goal is to demonstrate that NCT can now achieve a comparable spatial resolution to XCT at a thin-section level. A number of examples have been chosen to illustrate the complementary nature of the two methods. In particular, we focus on identifying hydrous phases using NCT and studying large samples with good levels of contrast and image quality.

METHOD

Compared to synchrotron radiation sources, even the most powerful neutron sources are rather weak (Banhart, 2008). For high-resolution imaging, a highly collimated beam is available only at very few imaging facilities, at reactor sources such as the Forschungs-Neutronenquelle Heinz Maier-Leibnitz (FRM-II) of the Technische Universität München, Germany; the National Institute of Standards and Technology (NIST), USA; and the continuous spallation source at Paul Scherrer Institute (PSI), Switzerland (Winkler, 2006). Since it is not possible to create a magnifying cone beam like an X-ray tube, parallel beam geometry is approximated by a small-diaphragm collimator and a long flight tube. The image resolution is thus limited by the detector, since there is no inherent image magnification. A neutron sensitive scintillation screen (ZnS + LiF) is commonly used in combination with a mirror and a scientific cooled CCD camera (Lehmann et al., 2007). One of the current limitations in resolution is given by the scintillation screen, due to an inherent blurring of the radiograph. Slow neutrons (defined as those with energies <0.4 eV) can only be detected by a nuclear reaction. The detection reaction 6Li + n → 3H + 4He is exothermic with an energy of 4.7 MeV. This energy is distributed as kinetic energy among the reaction products, the 4He and 3H particles. The reaction products are then stopped within the zinc sulfide (ZnS) scintillation material. The stopping path length of these particles blurs the location of each neutron event on the screen. For example, the average range in a standard 100-μm-thick screen is ∼50–80 μm (Spowart, 1969). Because gadolinium is very sensitive to neutrons, thinned X-ray scintillation screens made of gadolinium oxisulfide (Gd2O2S) can be used for neutron detection. At present, Gd2O2S screens are the only neutron-sensitive scintillators commercially available in the range of 10–20 μm thickness (Baechler et al., 2002). The blurring on such screens is reduced and they have lower capture energy; however, they produce less light output, leading to longer exposure times to achieve the same image quality. With 20-μm-thick screens, we can currently achieve roughly 20 μm image resolution.

The beam flux is a primary criterion on the exposure time necessary for each radiograph. This depends on the neutron statistics required to clearly distinguish between the material phases of interest. For example, a high contrast problem such as discriminating between regions of metal and air in an engine block requires a relatively low number of neutron counts. This means that, for some industrial applications, even a low-flux beamline can produce tomographic reconstructions in a timely fashion. However, for geomaterials the contrast between the phases of interest is often subtle. While it is possible to increase the flux by increasing the size of the pinhole, this reduces the spatial resolution of the reconstruction. Therefore, in order to avoid prohibitively long exposure times for high-resolution geomaterial NCT, we are restricted to beamlines that can simultaneously achieve high beam flux and collimation. To date, ANTARES is one of the very few such facilities available for applied research (Schillinger, 2009).

To study the properties of the new high-resolution scintillators we imaged a Siemens star and square grid test pattern (Grünzweig et al., 2007) (Fig. 3A). Radiographs were produced for different scintillator thicknesses (Fig. 3B), exposure times, and beamline configurations. We also tested the effect of geometrical blurring by increasing the distance between the test pattern and the scintillator screen (Fig. 3C). In Figure 3A, the mesh on the right side of the test pattern consists of sharp lines of three different widths (50 μm, 100 μm, and 150 μm). To determine the resolution of each radiograph, we measured the apparent full width at half maximum (FWHM) of each of these lines and compared this to the true line width. The FWHM was measured from the CCD measured intensity in a cross section of each line type in the radiograph.

In order to study the relationship between image quality, scintillator thickness, and exposure time we performed a second set of experiments. Radiographs of the test pattern were measured with different exposure times for a 20-μm-thick Gd2O2S scintillator. The exposures were carried out for two different beam collimations. Beam collimation is characterized by the ratio of beam flight length, L, to pinhole diameter, D, the L/D ratio. The higher the value of L/D, the more collimated the beam. In these experiments we used configurations of L/D = 400 and 800, both with and without a lead filter. We hoped that the lead filter would reduce the number of gamma events and thereby improve the overall image quality. To determine the relative image quality of each radiograph we measured the standard deviation of the radiograph intensity for one of the textures on the test image.

The XCT measurements were performed using an IZFP device (Institut für Zerstörungsfreie Prüfverfahren; Fraunhofer Society for the Advancement of Applied Research, Institute for Mineralogy, Crystallography and Material Science, University of Leipzig). Details and measuring parameters are given in the appendix.

RESULTS

The size of the blur spot on the radiograph is the combination of three main effects: (1) the geometrical spreading, which increases with distance to the scintillator screen; (2) the stopping length of the reaction products within the screen; and (3) a cutoff due to the CCD camera and lens resolution, as features smaller than the pixel size or lens resolution cannot be resolved on the radiograph.

Figure 4A illustrates these blurring components, and Figure 4B shows the measured resolution for each of the different scintillators (100 μm, 50 μm, 20 μm, 10 μm thickness). Thinner scintillators achieve better resolution. The 10 μm scintillator resolution may be better than indicated, as 16 μm was the CCD pixel size cutoff for these experiments. The total blur spot radius (rtot.) can be expressed as: 
graphic
where d is the distance from the test pattern to the scintillator screen. The coefficients rs and cG are given in Table 1. All units are in micrometers. These coefficients vary depending on the beam collimation, but should otherwise prove universal for different experimental facilities and conditions.

Figure 4B shows that geometrical spreading becomes significant for experimental scenarios with low beam collimation and high sample radius. The sample radius dictates the minimum possible distance of the rotation axis from the scintillator screen, d. The blur spot radius climbs more steeply as a function of d for lower values of L/D. For example, consider a sample of radius 70 mm imaged by a 20 μm scintillator. The blur spot size doubles from L/D = 800–400. At first glance this suggests that the best possible beamline setup is to use very thin scintillators and high beam collimation. However, we must also consider the image quality and exposure time for each radiograph. Image quality is a combination of two main effects. The first is from neutron statistics. In order to distinguish between different material phases, the CCD camera must detect a given number of neutron events. The number of events required depends on the contrast between the materials. For high contrast, this number is low; but for phases with subtle contrast, a very high number of neutrons is needed. A thin scintillator is less likely to detect a neutron than a thick scintillator.

The second effect is from so-called “gamma” events. High-energy X-rays and gamma rays strike the camera chip itself, resulting in a bright spot on the CCD image. These gamma rays may be scattered primary gammas from the reactor, or gammas created by neutron capture in the sample itself. The intensity of these gamma spots, as measured by the CCD, is in no way representative of the neutron attenuation of the material, so these spots must be discarded. In practice the corrupted pixels are replaced by interpolating between values of uncorrupted pixels in the neighborhood. This filtering reduces the effective spatial resolution of the image.

Figure 5A shows how the standard deviation decreases with exposure time. The following function represents the best fit to the experimental data, describing the standard deviation, σ, as a function of time, t, 
graphic
where σ0 is the inherent standard deviation of the texture and c is a constant. This depends on the beam flux and type of scintillator. Figure 5A shows that high collimation is achieved at the cost of beam flux. Therefore, a significantly longer exposure time is required to achieve the same image quality. The same is true when using the lead filter (Fig. 5B). Although the filter reduces the number of gamma events for a given exposure time, a longer exposure time is required to achieve the same standard deviation. When we plot σ versus gamma corruption (Fig. 5C), we find that the Pb filter does not significantly improve the overall image quality, as all curves lie roughly in the same region. The Pb filter can only reduce the primary gammas from the reactor, while the major part of the gammas thus seems to be generated in the sample or the structural materials of the detector.

Figure 6 shows that thinner scintillators require more exposure time to achieve the same image quality. The colored squares show measured radiographs, while the lines represent different values of c (equation 2). Table 2 shows the image noise coefficient for each scintillator and collimation. In Table 3 we show the exposure time required for an 800 step tomography with L/D = 800 and σ = 0.03. Based on the large visible jump between 50 μm LiF + ZnS and 20 μm Gd2O2S screens in Figure 6, we can confirm that the lower light output from Gd2O2S screens results in significantly longer exposure times to achieve the same image quality. We estimate the relative exposure time to be almost an order of magnitude higher for Gd2O2S, in agreement with previous work (Baechler et al., 2002).

State-of-the-art NCT imaging can now achieve a spatial resolution of <20 μm. This can be done using a combination of high beam flux, high collimation, and thin scintillator screens. However, this configuration results in either poor contrast or prohibitively long exposure times. For high contrast cases, such as between air and rock, this configuration is ideal. For cases where a more subtle contrast between phases is expected, thicker scintillation screens may be preferable. The optimum configuration must be determined on a case by case basis for each sample. High beam collimation is particularly important for larger samples, as the geometrical blurring quickly becomes the dominant effect at higher radii.

DISCUSSION AND APPLICATION

We demonstrate the feasibility and complementary nature of the method by showing a number of examples. These examples have been selected from work that is in progress and each will be presented in its entirety in future publications. In this paper we use them to demonstrate the feasibility and usefulness of NCT as a method for characterizing geomaterials. We show here that NCT can achieve a spatial resolution comparable to that of XCT at a thin-section level. In particular, we focus on NCT as a tool for the identification of hydrous phases and high-contrast imaging of large samples.

Example 1. Muong-Nong–Type Tektite

Tektites are natural glasses formed during meteorite impacts. They are found in four distinct strewn fields on Earth (North American, central Europe, Ivory Coast, and Australasian). The Australasian strewn field is the largest by far, covering approximately one-third of the Earth's surface. Tektites from the strewn field are commonly as large as a few centimeters, and have chemical and isotopic compositions that are very similar to those of terrestrial upper continental crustal rocks. Studies have demonstrated that they were formed during hypervelocity impacts on Earth. There are many unanswered questions regarding the formation process (Trnka and Houzar, 2002). Some tektites contain bubbles filled with gas that relates to the formation conditions (Fehr, 1999; Meisel et al., 1993). They occur in all types of tektites (Muong Nong or layered, splash-form, and aerodynamically ablated tektites on land, microtektites in deep-sea sediments). Muong-Nong–type tektites differ in appearance from splash-form tektites by having irregular blocky shapes and a layered structure. All previous studies ultimately led to the destruction of the samples (Fiske, 1996; Matsuda et al., 2000). Only one study applied a nondestructive method (high-resolution X-ray computed tomography; Koeberl et al., 2002) on tektite samples. However, this was conducted at a micrometer scale and the major part of the work focused on impactites (impact breccias).

Very few data are available on the size and spatial distribution of bubbles and the chemical layering in tektites. Tomographic techniques can deliver large-scale (centimeter range) 3D distribution patterns with reasonable resolution (50 μm) without destroying the tektite samples. Therefore, the scope of this application is twofold: to determine the size, shape, and spatial distribution of bubbles and chemical layering (1) within a single tektite by means of a nondestructive method; and (2) in different tektites of different parts of a single strewn field. The final results should enable us to constrain the formation processes of different types of tektites from the Australasian strewn field.

Figure 7 is a comparison between NCT and XCT imaging for a Muong-Nong tektite. These overview scans will allow us to select regions of interest for drilling and future high-resolution studies. The NCT reconstruction was performed using a 100 μm scintillator. Further experimental details, for this and other samples, can be found in the Appendix. In Figures 7A and 7B, brightness increases with attenuation. The third image (Fig. 7C) is a false-color combination of the two data sets, with the neutron data in the red channel and the X-ray data in the blue channel. Both data sets share the green channel equally. Where the two data sets agree on the relative attenuation, the combined region will be shaded gray; if neutron or X-ray attenuation dominates, the region will be tinged orange or blue, respectively. We can distinguish between the unaltered glass matrix (blue), deformed pores (black), and hydrogen-bearing material (orange). We speculate that the hydrogen-bearing material is mud that has been entrained into the tektite during landing, partially filling the pore space. Some of this muddy material is still visible on the surface of the sample. We speculate further that this entrainment process occurred at high temperatures. This would lead to the release of water from the mud into the pristine glass matrix, resulting in water diffusion profiles.

Example 2. Seismogenic Lavas

Understanding the ductile-brittle transition in dome lavas may well contain the key to an adequate description of dome growth and stability. To elucidate this transition in dome lavas, a series of experiments was performed to characterize microcracking during compressive deformation of crystal-rich lavas (Lavallée et al., 2007). Multiphase lavas behave as viscoelastic fluids with a strain-rate dependence on viscosity across the ductile-brittle field. The rheological results, coupled to the microseismic signal, support the association of seismic swarms with seismogenic shear zones during eruptions. Given our observation that lavas may behave like their volcanic rock equivalent at high strain rates, careful monitoring of their seismicity could be coupled to failure forecast methods to successfully predict impending lava dome eruptions in a volcanic crisis (Lavallée et al., 2008).

Acoustic emission and postexperiment textural analyses on pristine, unaltered dome rock samples reveal that the crystals, in particular, appear to influence crack propagation, and in some cases strong, intact crystals may deviate crack propagation. Moreover, there was a tendency for the plagioclase crystals to be partially fragmented during the deformation (Cordonnier et al., 2009).

Directional drilling at the Unzen Volcano, Japan, in 2004 penetrated through the 1991–1995 magma conduit and successfully recovered samples of the lava dike. The dike was sampled ∼1.3 km below the volcano's summit vent and intruded into a broader conduit zone that is 0.5 km wide. Unexpectedly, the lava dike sample was altered, suggesting that circulation of hydrothermal fluids rapidly cools the conduit region of even very active volcanoes (Kusakabe et al., 1999).

Figure 8 shows a scientific drill core taken from Mount Unzen. It has been taken from a hydrothermally altered part of the volcanic conduit. In Figure 8 we show three perpendicular cross sections of the drill core taken with XCT and NCT. The contrast between phases for the NCT image is noticeably better than XCT. The XCT reconstruction also has strong image artifacts. This is a well-known difference between the two methods. When using polychromatic neutron and X-ray spectra, the neutrons usually undergo less beam hardening than XCT. Beam hardening only becomes significant for NCT investigations involving either neutron absorber materials such as boron, dysprosium, or gadolinium, or a high content of water (Vontobel et al., 2006). This artifact makes false-color compositions difficult to interpret. Hydrothermal alteration is visible in the boxed region. In Figure 8A a faint ring-like shadow is visible in the image, possibly indicating a thin crack or grain boundary. The same feature in the NCT reconstruction (Fig. 8B) attenuates very strongly, suggesting the presence of hydrogen as water (OH, H2O) incorporated in the newly developed mineral phases. The Mount Unzen drill core is decorated with many such features. NCT provides the first opportunity to nondestructively map and study hydrothermal alteration and diffusion in three dimensions.

Example 3. Tooth Enamel Microstructure in Fossil Hominids

Tooth enamel microstructure is an important archive for taxonomic, developmental biologic, physiologic, ecologic, and environmental information in paleontology and in particular paleoanthropology. Hominoid fossils (human relatives) are usually preserved as isolated teeth, because teeth are the most easily preserved parts of a vertebrate body. Tooth enamel microstructure analysis is therefore crucial in studying hominid evolution and life histories (Olejniczak et al., 2008). The hard tissues of mammalian (and so hominoid) teeth are composed of dentin and enamel. Paleobiological information in hominoid molars is preserved in the thickness and distribution of enamel, the morphology of the dentine horns, as well as the enamel microstructure. During enamel formation, hydroxyapatite crystallites are bound together in long, thin enamel prisms, representing accretionary products of enamel-forming cells. Enamel prisms preserve a record of daily secretion, represented by circadian features termed cross-striations and/or laminations (Tafforeau et al., 2007). In order to accurately assess tooth crown formation, previous studies of intact teeth have employed physical sectioning of tooth crowns to visualize the key relationship between cross-striations and successive positions of the forming front known as Retzius lines (Smith et al., 2007). The 3D enamel thickness, especially with regard to aspects of enamel-dentine junction morphology, is an important taxonomic tool to differentiate between hominoid phylogenetic lineages.

The goal of the study is to employ new nondestructive 3D tomographic techniques to examine the microstructure, thickness, and distribution of enamel in molars of recent humans and two fossil pongins (orangutan relatives) Gigantopithecus and Sivapithecus and to examine aspects of enamel-dentine junction morphology (e.g., the height of dentine horns) as they relate to enamel thickness. In addition, we plan to explore the potential of this technique in visualizing the 2D enamel, key components in assessing age at death in fossil dentitions. This will be especially important for the planned investigation of the youngest European fossil great ape, found recently in Bulgaria (Spassov, 2008), for which this investigation will be the pilot study.

The sample shown here is much smaller than the previous examples and high spatial resolution was required to see the details of interest. Figure 9 shows a human tooth. The NCT has been done with a 10 μm scintillator. Note that the image noise is higher than in other examples; this is due to the relatively low light output of the thin Gd2O2S scintillator screens. In order to achieve a comparable image quality, a much longer exposure time was required per radiograph. In this example a suboptimal exposure time was used due to beam-time limitations. Once again, the complementary nature of NCT and XCT imaging is very striking. Neutrons are strongly absorbed by the core (dentine) material, while X-rays traverse the same region with relative ease; this is due to the high water content of dentine with respect to enamel (Table 4) (Goodman et al., 2004).

CONCLUSIONS

NCT provides complementary information to XCT imaging. The method's sensitivity to hydrogen is particularly interesting to Earth scientists as this allows us to track the presence of water (OH or H2O) within a sample, and to distinguish easily between hydrous and nonhydrous mineral phases. Until recently, the method has not had extensive application in general research due to its limited availability and spatial resolution (∼200 μm). Hardware improvements at the ANTARES beamline have significantly improved NCT as a tool for the Earth sciences, allowing us to achieve spatial resolutions of <20 μm using thin scintillator screens and high beam collimation. There is an inherent tradeoff between good resolution, exposure time, and image quality. Higher exposure times are required for thin scintillators to achieve the same image quality. Increasing the exposure time also increases the fraction of radiograph pixels corrupted by high-energy X-rays and gamma rays. While it is possible to remove these gamma pixels using a computer algorithm, doing so effectively reduces the spatial resolution of the image. For large samples it is crucial to use high beam collimation, as low collimation geometries quickly lead to unacceptable levels of geometrical blurring. The optimum configuration must be determined for each sample on a case by case basis.

With its greatly improved spatial resolution, NCT now promises to be a useful tool for a much wider spectrum of geomaterial research. It can be used to study the formation of natural glasses and for the characterization of precious samples from scientific drilling, where shear zones are often decorated with hydrous minerals. Such features can be found, for example, in samples from the San Andreas Fault Observatory at Depth (SAFOD) and the Kontinentales Tiefbohrprogram der Bundesrepublik Deutschland (KTB), or the hydrothermally altered ascent path for volcanic conduits, e.g., the Mount Unzen Scientific Drilling Project (USDP). Visualization of stepwise rock deformation experiments allows us to observe the development of shear zones under controlled conditions. The availability of samples, both from the laboratory and the field, promises to shed light on the process of high-temperature rock cracking and magma failure. Nondestructive tomographic characterization of molar sections allows the full 3D distribution of enamel thickness to be examined, rather than a single section plane. This may help resolve the debate over life-history differences between fossil hominids, living great apes, and humans.

APPENDIX. EXPERIMENTAL DETAILS

The NCT measurements were performed using the ANTARES beamline (FRM-II). This facility can provide geometries of L/D = 400 and 800 with neutron fluxes of 9.3 × 107 and 2.3 × 107 cm−2s−1. The acquisition is performed using a thermoelectrically cooled (–60 °C) CCD camera (ANDOR DW436) with 2048 × 2048 pixels and a Leica APO Macro Elmarit-R 100 mm f/2.8 lens. Table 5 summarizes the experimental conditions used for each sample.

The XCT measurements were performed using a micro-CT device at the Institute for Mineralogy, Crystallography and Material Science at the University of Leipzig (Fraunhofer IZFP) with the following basic components: a microfocal X-ray source (by Feinfocus); an open tube system in transmission with a high-energy target (tube voltage of 150 kV, tube current of 100 μA); a seven-axis precision manipulator system (by TISAD, Technischer Industrieservice und Automation Dresden); and a 2048 × 2048 pixel flat panel detector (by Perkin Elmer). Table 6 summarizes the experimental conditions used for each sample.

We are grateful for the scientific discussions with and samples provided by R. Hochleitner and T. Fehr (Tektite); Y. Lavallée and S. Nakada (Mount Unzen drill core); and A. Rocholl and M. Böhme (human tooth). Financial support was provided for A. Flaws, K.-U. Hess, B. Schillinger, and D.B. Dingwell by the DFG-ICDP (Deutsch Forschungsgemeinschaft–International Continental Scientific Drilling Program) grant HE 4565-2-1.