Imaging in 3D under pressure: a decade of high-pressure X-ray microtomography development at GSECARS
- First Online:
- Cite this article as:
- Yu, T., Wang, Y. & Rivers, M.L. Prog. in Earth and Planet. Sci. (2016) 3: 17. doi:10.1186/s40645-016-0093-6
- 801 Downloads
The high-pressure X-ray microtomography (HPXMT) apparatus has been operating at the GeoSoilEnviroCARS (GSECARS) bending magnet beamline at the Advanced Photon Source since 2005. By combining the powerful synchrotron X-ray source and fast switching between white (for X-ray diffraction) and monochromatic (for absorption imaging) modes, this technique provides the high-pressure community with a unique opportunity to image the three-dimensional volume, texture, and microstructure of materials under high pressure and temperature. The ability to shear the sample with unlimited strain by twisting the two opposed anvils in the apparatus allows shear deformation studies under extreme pressure and temperature to be performed. HPXMT is a powerful tool for studying the physical properties of both crystalline and non-crystalline materials under high pressure and high temperature. Over the past 10 years, continuous effort has been put into technical development, modifications to improve the overall performance, and additional probing techniques to meet users’ needs. Here, we present an up-to-date report on the HPXMT system, a brief review of some of its many exciting scientific applications, and a discussion of future developments.
KeywordsHigh pressure Tomography Synchrotron Density Rotation Shear Non-crystalline Elasticity
Advanced Photon Source
cerium-doped lutetium aluminum garnet
diamond anvil cell
finite element method
high-pressure X-ray microtomography
lattice Boltzmann method
region of interest
Super Photon Ring- 8 GeV
yttrium aluminum garnet
Advances in high-pressure science rely critically on continuous technical developments and breakthroughs for new probes to characterize samples in situ under extreme conditions. With each new probing technique added to the tool kit, high-pressure research expands into new areas. Researchers have embraced every emerging new technique and have eagerly applied the techniques to scientific problems beyond the imagination of the original inventors. Literature shows that this has occurred throughout the history of high-pressure science. Excellent reviews highlighting the importance of new technical breakthroughs can be found in Bassett (2009) for the diamond anvil cell (DAC) and Liebermann (2011) for the multi-anvil press.
The development of three-dimensional (3D) tomographic imaging under high pressure and temperature began at the turn of the 21st Century at GeoSoilEnviroCARS (GSECARS) of the Advanced Photon Source (APS; Argonne National Laboratory, IL, USA). In 2005, a high-pressure X-ray microtomography apparatus was commissioned (Wang et al. 2005). The 3D tomographic imaging capability based on X-ray absorption was a major step toward investigating complex materials under high pressure and temperature. Soon after the commissioning of high-pressure X-ray microtomography (HPXMT) at GSECARS, high-pressure tomography began at The European Synchrotron Radiation Facility (Bromiley et al. 2009) and Super Photon Ring-8 GeV (SPring-8) (Urakawa et al. 2010) for a large-volume press and at APS for the DAC (Liu et al. 2008). Since then, high-pressure imaging techniques have expanded from absorption tomography to diffraction (Alvarez-Murga et al. 2011) and beyond.
Since the inception of HPXMT apparatus, numerous developments have been made at GSECARS to improve the mechanical performance of the apparatus, expand the pressure and temperature capabilities, increase imaging quality, and reduce data collection time. This paper reviews major technical improvements made over the past decade to the HPXMT apparatus, highlights the scientific-rich applications, and discusses possible new directions for future developments.
Basic setup of absorption-based microtomography
Pressure, temperature, and strain environment in the HPXMT apparatus
Improving load capacity of the apparatus
The load supporting columns were originally straight cylinders, the diameter of which expanded under high load, causing them to bind to the HarmonicDrive™ fitting units from time to time. Frequent servicing was required to clean the interfaces and replace lubricant. Therefore, a modification was made to the columns (Fig. 2b, c). The new columns have a slightly smaller general diameter, with only three circular lines matching the inner diameter of the HarmonicDrive™ fitting units, to ensure axial alignment of the system. This has completely eliminated potential binding between the columns and the fitting units. In addition, two circular grooves were added to the column profile to alleviate potential gumming when lubricant hardens. After this modification, the system has been used routinely up to 50 t (the designed load limit) and we have not seen any binding since. The mechanical strength and stability of the entire system have also been enhanced.
Pushing pressure capability
The pressure generating mechanism of the HPXMT apparatus is based on an opposed anvil design because of the need to have complete open access in the plane perpendicular to the vertical loading axis. Either Drickamer (Balchan and Drickamer 1961) or toroidal (Khvostantsev et al. 1977; Morard et al. 2007) devices can fulfill this requirement. The two anvils are pressed toward each other along the uniaxial load direction, with a sample assembly in between. Loading is provided by the 250 t hydraulic press installed at 13-BM-D of the GSECARS sector. This press is used for a number of high-pressure devices, such as the deformation DIA and the T-cup apparatus. Each device is mounted in a designated die-set unit with wheels fitted in the rail system mounted on the press frame (Wang et al. 2009). This allows easy switchover between different devices to suit users’ needs.
In HPXMT experiments, the temperature is estimated based on the temperature-power relation, which was established in separate experiments using a thermocouple. We find that the temperature-power relations vary with applied ram load, possibly reflecting deformation in the heater as well as changes in the distance between the heater and the anvil tips (Fig. 6a). This introduces some uncertainties in temperature measurements. A series of calibration curves is needed to obtain reliable temperature estimates. Alternatively, several metal foils (Ag, Au, etc.) with well-known melting curves can be placed in a single cell with in situ imaging to detect the onset of melting by monitoring the shape change in the foils, providing temperature fixed points (Clark et al. 2013).
Controlled deformation in HPXMT apparatus
System setup and upgrades
Imaging setup and upgrades
The HPXMT setup utilizes the synchrotron bending magnet beam (13-BM-D) at the APS. A Si(111) monochromator is used to select monochromatic radiation from 7 to 65 keV. Typical photon energies used for the HPXMT apparatus range from 25 to 45 keV. Initially, we used yttrium aluminum garnet (YAG) as scintillator crystals (0.3 mm thickness). This is adequate for lower energies, but at higher energies, the beam penetration is deeper in the scintillator. The depth of focus of the objective lens cannot discriminate signals from different depths of the scintillator, resulting in poorer resolution. Switching to CdWO4 scintillators improved image sharpness somewhat. Currently, we use single-crystal cerium-doped lutetium aluminum garnet (Lu3Al5O12) or Ce:LuAG (0.1 mm thickness). Because Lu has a much higher atomic number than Y, this screen has a higher stopping power. Rivers et al. (2010) have evaluated and compared the image quality between the YAG and Ce:LuAG scintillator crystals and shown that the image generated by the Ce:LuAG crystal is notably sharper.
Our imaging cameras have gone through several generations. Currently, we use Point Grey CMOS Model GS3-U3-23S6M, which has a 1920 × 1200 pixel array with a dynamic range of 73 dB. It has a maximum speed of 162 frames per second with a peak quantum efficiency of 76 %.
New integrated press, detector, and imaging position system
A narrow frame has been added between the press and detector manipulation systems for the imaging camera and scintillator assembly. This provides a more stable support for the imaging system and allows sample-scintillator distance to be freely adjusted, making it possible to apply phase contrast imaging technique to the HPXMT setup in the future.
The load feedback function in the 250 t press is less favorable during high-temperature data collection; the sample height may change because of the uniaxial nature of the opposed anvil devices, as the hydraulic ram compresses the HPXMT module along the vertical axis. This results in blurry tomography images. We have implemented a precision linear-voltage displacement transducer and added an option for feedback on the position of the hydraulic ram cylinder. This option keeps the sample cell vertical movement to a minimum throughout the tomography data collection.
Integrating additional probe techniques
Ultrasonic elastic wave velocity measurement has been implemented in the HPXMT setup (Kono et al. 2011). Figure 4a shows the location of the acoustic transducer for acoustic travel time measurement. Sample length is conveniently measured by X-ray imaging. This is extremely useful especially for studying non-crystalline materials.
Data collection, processing, and image reconstruction
New on-the-fly data collection mode
In the past, the collection was done in a stop-and-scan mode, with the rotation starting and stopping at each angle. The overhead was typically about 0.2–0.3 s at each point, depending on the motor velocity and acceleration values used. Not only did it increase the total data collection time, but also the frequent abrupt start and stop motion was also unfavorable for the mechanics of the HPXMT apparatus, especially when the system was under high load. To eliminate the start/stop overhead, Rivers et al. (2010) implemented on-the-fly data collection, with continuous motion of the rotation stage. Now, the motor pulses from the stepper motor controller are fed into an SIS3820 multichannel scalar, which divides the motor pulses by the number of motor pulses per rotation angle increment (N). The speed of the rotation stage is set so that the stage moves by 1 angular step (e.g., 0.25°) in the time required for the desired exposure plus the detector readout. If the optimum number of projections [(π/2) × NX, NX is the number of horizontal pixels)] is acquired, then the outermost part of the sample moves less than 1 pixel during the time to collect one radiograph in this continuous motion mode. This development has greatly improved the mechanical stability of the HPXMT system.
Dark current and flat field normalization
The raw data collected with the Point Grey camera are 12 bit 1920 × 1200 pixel images. The first correction that must be applied to such images is for dark current and flat field. Dark current is the signal recorded in the absence of X-rays. We typically take multiple 5 s exposures to obtain an average dark current. The flat field is the image that is measured with the X-rays on, but without a sample in the beam. The non-uniformity in the flat field includes effects of non-uniformities in the incident X-ray beam and non-uniform responses of the scintillator and the imaging detector. In ambient tomography, the sample is surrounded by nothing except for air. In HPXMT, the sample is surrounded by a capsule, a heater, and layers of pressure medium, plus the containment ring. In order to maximize pixel coverage on the sample for better spatial resolution, we use the pressure medium as the intensity background instead of air. Flat field images are taken through a dummy cell assembly, which has identical pressure medium to the true sample assembly, but with the sample removed. In practice, the dummy is mounted ~20 mm away from the high-pressure assembly; flat field images are taken by driving the hydraulic press so that the dummy is in the X-ray path.
The speed of image reconstruction process was significantly improved by using the new tomoRecon multi-threaded code (Rivers 2012). This code uses the high-speed Gridrec FFT algorithm and reconstructs N slices in parallel on N cores in the workstation. The time to reconstruct a 1920 (X) × 1200 (Y) × 900 (projections) dataset is now less than 60 s. The reconstruction code is written in C++ as an IDL user interface. If needed, the reconstruction code can be run free of cost by users in their home institutions.
Rotation center correction
The high-pressure nature of HPXMT apparatus makes the rotation center mechanically less accurate than conventional tomography with high-precision motors. The poor centering causes artifacts in the tomography images. During data processing, the software determines the “center-of-gravity” of each row in each sinogram and fits this center-of-gravity array to a sine wave. This results in a sinogram which is centered better on the rotation axis. The sinogram is then shifted left or right so that the rotation axis is located on the center column of the sinogram array. In cases where this automatic centering function is insufficient for the correction, a manual tweak function is available to tweak the sinogram until the resulting reconstruction is satisfactory.
Image analysis and manipulation after tomography reconstruction are conducted using free software such as ImageJ (http://imagej.nih.gov/ij/) and Blob3D (Ketcham 2005; http://www.ctlab.geo.utexas.edu/software/blob3d/). The tool kits included in these software packages allow texture analyses, strain calculations, and volume calculations. Especially for ImageJ, numerous special tools have been created by its user community and many are available online free of charge. Sophisticated users use commercially available software packages that have the specific functions they need.
A wide range of scientific projects have used high-pressure X-ray microtomography. Topics cover subjects in many fields such as physics, materials science, earth science, and civil engineering. In this section, we highlight recent applications that have made a significant contribution to the corresponding field.
Density of non-crystalline material under pressure
Volumetric rendering was also applied to solid and liquid Ga to study its compression behavior (Li et al. 2014). The rapid switching between monochromatic (for tomography imaging) and white mode (for energy-dispersive diffraction) made it convenient to identify crystalline phases and melting. This technique was described by Lesher et al. (2009). Two isothermal compression curves were obtained on liquid Ga at 300 and 330 K up to 3.6 GPa (Fig. 9c). Above 2.44 GPa, the 330 K curve showed a significant bend-over, which was interpreted as due to a liquid–liquid transition.
Measuring interfacial tension by imaging liquid droplets
A novel application of the volume rendering measurement in HPXMT is to measure interfacial tension between liquids under high pressure and temperature. The fundamental principle is the well-known Laplace pressure caused by the surface tension or energy of the interface between liquid and gas. At 1 atm, a liquid (L1) drop sitting on the flat surface of a solid in another liquid (L2; or gas such as air) takes a certain shape, which is determined by the balance of the surface energies between L1 and L2 (γL1 − L2), liquid and solid (γL1 − S), and solid and air (γS − L2) in the presence of a density difference (∆ρ) between L1 and L2. This balance is shown by the contact angle (θC) between the drop of L and the solid plate. The technique based on this principle to measure interfacial tension is called the sessile drop method (Rotenberg et al. 1983), which shows that if the drop is axisymmetric (with the symmetry axis parallel to gravity), surface tension between liquids L1 and L2 can be determined by measuring the shape of the liquid drop. Terasaki et al. (2008, 2009) extended this technique to high pressures and measured the surface tension of liquid Ni-S, Fe-S, and Fe-P in liquid sodium disilicate by using HPXMT. A fully densified alumina plate was used in the high-pressure cell assembly for the sessile drop. They found that at 1.5 GPa, the interfacial tension of liquid Fe-S decreased significantly from 802 to 112 mN/m with increasing sulfur content from 0 to 40 at.%. This effect of 40 at.% S on interfacial tension was large compared with the effect of temperature (~273 mN/m reduction with an increase of 200 K). In contrast, interfacial tension in the Fe-P system increased only slightly from 802 to 873 mN/m as phosphorus content increased to 17 at.%. Both tendencies are consistent with those measured at ambient pressure on Fe-S and Fe-P liquids.
Only when the shape of the drop is axisymmetric can the sessile drop method be used for interfacial tension measurement. HPXMT is critical in checking the shape of the drop in situ at high pressure (Fig. 9d, e). When a melt is quenched, the shape of the drop changes suddenly, resulting in drastically different apparent interfacial tension values, if the quenched shape is used. Furthermore, it must be ensured that the melt is compositionally homogeneous. Because Fe and S (or P) have very different densities and mass absorption coefficients, any segregation of Fe and/or S (P) in the melt can be easily observed (Terasaki et al. 2009).
3D imaging of microstructural heterogeneities
The power of HPXMT apparatus lies in the ability to resolve small heterogeneities under high pressure and temperature and various strain conditions. This allows in situ tracking of various components in complex materials, which is required in many scientific disciplines. Below are a few examples highlighting the great potential of this technique.
Shear-induced fabric transition in multiphase composites
Shear deformation in metal-silicate composites and applications to core formation processes in planetesimals
Bulk-proppant packs and propped fractures under uniaxial load
Compression mechanism of silica glass: Combining acoustic velocity measurements with HPXMT
Imaging-based volume/density measurements of non-crystalline materials are intrinsically less accurate, with typical errors on the order of 2 %. This is more than ten times the uncertainty in volume/density measurements of crystalline materials based on X-ray diffraction. To characterize equations of state of non-crystalline materials more accurately, Kono et al. (2011) combined HPXMT imaging with ultrasonic velocity measurements. An ultrasonic transducer can be easily mounted on the back side of the PE anvil (e.g., Fig. 4a), without being compressed during high-pressure experiments. Ultrasonic measurements require larger sample diameters in order to obtain strong acoustic signals, and the sample length should also be large enough to avoid overlapping in the acoustic signals reflected at both ends of the sample. In addition, the use of large samples is advantageous for precise measurement of sample volume, although the maximum attainable pressure would be somewhat limited. A SiO2 glass disk ∼2.5 mm in diameter and 0.429 mm in length was loaded in the cell assembly. Au foils (2.5 μm thick) were placed at both ends of the sample, serving as boundary/interface markers for length measurements using X-ray imaging. At each pressure point both P- and S-wave travel times were measured at 20 and 30 MHz for S- and P-waves, respectively. X-ray tomography provided direct volume measurements and sample length, which was then used to calculate P- and S-wave velocities from the travel times. By assuming that the glass sample remained elastically isotropic at high pressures, the sample density could be also obtained by integrating the velocities over the pressure range (Kono et al. 2011). This yields a set of density data that is completely independent of sample volume measurements.
Prior to the experiment, the density of the silica disk was determined by Archimedes’ method to be 2.208 g/cm3. This was used to distinguish various thresholding methods in determining the ambient sample volume in HPXMT. It was determined that the Gaussian filter gave the most consistent volume with ambient density data. By adopting the Gaussian filter, relative density changes of the sample at high pressures were determined from volume rendering.
The two density data sets, from ultrasonic velocities and tomographic volume measurements, showed excellent agreement. This allowed further determination of the bulk (Ks) and shear (G) moduli of silica glass. Both Ks and G decrease with increasing pressure up to ~2.5 GPa, above which Ks and G began to increase again. This softening behavior is related to the polymerized nature of the corner linking SiO4 network in silica, a topic of great interest in the liquid and glass community (e.g., Vukcevich 1972; He and Thorpe 1985). This work demonstrates the power of combining acoustic velocity measurements with HPXMT in studies of the compression behavior of non-crystalline materials. Velocities are directly related to the bulk modulus, which is the pressure derivative of volume. With information on both volume and its pressure derivative, more precise equations of state can be established.
The HPXMT setup at GSECARS is a powerful in situ imaging tool, serving the high-pressure community with various projects that are of interest to a wide range of scientific disciplines. The system now runs routinely up to 8 GPa in pressure and 2000 K in temperature. Many improvements and new developments have been undertaken, as described in the paper. On the 10th anniversary of the HPXMT technique, several promising future prospects can be envisioned.
1. Exploring new anvil materials and geometries will allow us to extend the pressure capability significantly. The CDT anvils are promising for up to 20 GPa. Utilizing sintered diamond as an anvil material will push the pressure limit further. New heater materials will be explored for higher temperature experiments. It may be necessary to upgrade the existing HPXMT apparatus to greater load capacity for higher pressures.
2. For parallel-beam absorption X-ray tomography under pressure, the presence of a pressure medium surrounding the sample presents a challenge for high-resolution imaging. Phase contrast tomography can be readily performed in the existing HPXMT apparatus, as the sample-detector distance can be easily adjusted with the integrated nine-axis press/detector manipulation system. This may help improve image quality in some systems. Another option to improve resolution is to combine conventional tomography with known-subregion-based interior tomography (Wang and Yu 2013). Mathematically speaking, tomography produces the reconstruction of a function F from a large number of line integrals of F. Conventional tomography is a global procedure because the standard convolution formulae for reconstruction at a given point require the integrals over all lines within the plane containing that point. In contrast, interior construction is a technique to correct truncation artifacts caused by limiting image data to a small field of view. The reconstruction focuses on an area called the region of interest (ROI). The problem of non-uniqueness in interior reconstruction may be solved by global reconstruction at relatively low resolution to facilitate higher resolution interior reconstruction.
3. In addition to absorption tomography, the setup can be readily extended to other 3D imaging techniques such as X-ray fluorescence tomography of high atomic number elements for chemical composition mapping and diffraction tomography for structural mapping.
4. We have seen an increasing demand for low-pressure (up to several hundred MPa) tomography, which requires separate apparatus. We envision the new apparatus to have axial deformation capability, with options for pore fluid and pore pressure control.
We thank T. Uchida for his contribution to the earlier phase of the HPXMT development and F. Westferro for his excellent engineering support. We thank our users for their valuable input, especially C. Lesher, A. Clark, R. Li, Y. Kono, H. Watson, and C. Wilson for their contributions to the science highlights summarized here. The presented experiments were performed at GSECARS (Sector 13), APS, Argonne National Laboratory. APS is supported by the US Department of Energy, Office of Basic Energy Sciences, under Contract No. DE-AC02-06CH11357. GSECARS is supported by the National Science Foundation - Earth Sciences Program (EAR-1128799) and by the Department of Energy - Geosciences (DE-FG02-94ER14466). We are grateful for the generous support of the National Science Foundation (EAR-0001088, 0711057, and 1214376) which made this development possible.
TY and YW set up the experimental hardware and composed the manuscript. MR wrote the tomography data collection, processing, and image reconstruction software. All authors have read and approved the final manuscript.
The authors declare that they have no competing interests.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.