Characterization and Calibration of the CheMin Mineralogical Instrument on Mars Science Laboratory

CheMin has a microfocus X-ray tube with a Co anode and a beam-defining final aperture placed at some distance from the tube, which together produce a collimated X-ray beam. The source produces continuum and characteristic X-radiation from a 50 μm diameter spot on the Co anode. The tube has a nominal operating voltage of 28 keV with a filament current of 1.5 A and cathode output of 100 μA. The photons emanate from the X-ray tube in a cone of radiation that exits the tube through a beryllium window. At a specific distance from the tube, the pinhole aperture intersects the beam, creating a nearly parallel, collimated ∼70 μm diameter beam of X-radiation directed at the center of the sample cell. After passing through the sample, the direct beam is stopped by a beam trap mounted on the edge of CCD detector.

The X-ray power supply is enclosed in a housing pressurized with a mixture of sulfur hexafluoride and nitrogen gas, mixed so that the condensation temperature of the gas is below the minimum expected temperature in the body of the rover (to ensure that the gas never liquefies).

The CheMin sample handling system consists of a funnel, a sample wheel (which carries 27 reusable sample cells and 5 permanent reference standards), and a sample sump where material is dumped after analysis (Fig. 2). CheMin receives sieved and portioned drill powders and scoop samples from the SA/SPaH system (Jandura 2010) through the CHIMRA drill, scoop, and sorting assembly (Sunshine 2010). A maximum of 76 mm³ of sample material is delivered to the piezoelectrically vibrated funnel system that penetrates through the rover deck. When CheMin is not receiving samples, the CheMin inlet is protected by a cover. The material received through the funnel passes into a sample cell that consists of a 10 mm³ active cell volume and a ∼400 mm³ reservoir above the cell. Excess material, should it be delivered, will remain in the reservoir during analysis. The funnel contains a 1 mm mesh screen to keep larger grains from entering the CheMin sample handling system. Grains that cannot pass through the screen will remain there for the duration of the mission, although no material is expected because the sample is pre-sieved to <150 μm in the CHIMRA sorting chamber to prevent clogging of the CheMin funnel screen. Any grains smaller than 1.0 mm but larger than 150 μm will pass into the upper reservoir portion of the sample cell, where they will remain until the cell is inverted and they are dumped into the sump. For the lifetime of the mission, nominally one Mars year, CheMin is required to accept and analyze material delivered from SA/SPaH with no more than 5 % internal contamination between samples. Self-generated contamination may originate from material that has remained in the funnel from previously delivered samples (and delivered along with subsequent samples), or from material that has remained in previously used analysis cells (CheMin will be capable of reusing each cell two to three times to accommodate 74 or more analyses during an extended mission). Cells are emptied after an analysis by rotating the sample wheel 180^∘ (to invert the cell) and vibrating the cell so that the sample material is emptied into a sump at the bottom of the instrument. If contamination is suspected either from the funnel or from a previously used cell, CheMin can reduce sample-to-sample contamination by dilution. Aliquots of sample material can either be dumped into the funnel and delivered directly to the sump through a shunt in the wheel without entering a sample cell (to entrain and remove funnel contamination), or a previously used sample cell can be filled, shaken and emptied to the sump prior to receiving a second aliquot of sample for analysis (to entrain and remove sample cell contamination). These processes will require coordination with SA/SPaH to deliver more than one aliquot of a given sample.

Open image in new window

The collimated ∼70 μm diameter X-ray beam illuminates the center of a sample cell having 6 μm thick Mylar^® or 10 μm Kapton^® windows. The sample introduced into the funnel consists of ≤76 mm³ of powdered material with a grain size of <150 μm. Only about 10 mm³ of material is required to fill the active volume of the sample cell, which is a disc-shaped volume with an 8 mm diameter and 175 μm thickness. The remaining sample material occupies the reservoir above the cell (see Fig. 3). During filling, analysis and sample dumping, the sample cell is shaken by piezoelectric actuators (“piezos”). The modes in which the piezos will be driven are still under test and may vary from sample to sample, depending on sample-dependent characteristics such as grain cohesion (e.g., clay-rich samples versus samples that lack fine particles). In CheMin testbeds and prototype instruments, the frequency of the piezo-actuator is varied so that during a part of the cycle the sample holder is at resonance, at which time the sample exhibits bulk convective movement similar to a liquid, delivering sample grains in random orientation into the volume exposed to the beam. In the CheMin Flight Model (FM) and the equivalent Demonstration Model (DM) a nominal resonant frequency of 2150 Hz is maintained during analysis, and shaking amplitude is varied to adjust the intensity of grain motion.

Open image in new window

During the moderate shaking that produces granular convection, it is possible that phase separation will occur as a result of size, density or shape differences between individual mineral grains. To mitigate this problem CheMin at intervals uses episodically larger shaking amplitudes (i.e., “chaos mode”) to re-homogenize the material in the sample chamber.

The CheMin sample cells are paired in dual-cell “tuning-fork” assemblies with a single horizontally driven piezoelectric actuator in each assembly (Fig. 3). Sixteen of the dual-cell assemblies are mounted around the circumference of the sample wheel (Figs. 2 and 4). Five of the cells will be devoted to carrying standards; the other 27 cells are available for sample analysis and can be reused by emptying samples into the sump after analysis. Cells are filled and analyzed at the top of the wheel. A cover assembly sits directly above the sample cell during analysis to keep material in the sample reservoir from being ejected during vibration. Inlets to the cells holding standards are sealed with HEPA filters to prevent the material from falling out as the wheel is moved, while allowing for pressure equalization within the cells.

Open image in new window

Both Mylar^®- and Kapton^®-windowed cells are mounted on the wheel. The rationale for using two types of cell windows is based on our experience with CheMin prototype instruments. Mylar^® windows exhibit a very flat diffraction background across the full range of 2θ. However Mylar^TM is less durable than Kapton^® under severe vibration and is susceptible to destruction if highly acidic samples (e.g., copiapite; a hydrated iron sulfate) are encountered. What is known about mineralogy on Mars suggests that acidic conditions are not unlikely (e.g., determination of jarosite occurrences at both MER landing sites). Kapton^® windows are more durable and are not susceptible to acid attack, but they have a small diffraction contribution at ∼6–7^∘ 2θ that could interfere with the 001 diffraction peak from some clay minerals. For security and to assure that CheMin can handle a wide range of sample types on Mars, both Kapton^® windows (in 13 cells) and Mylar^® windows (in 14 cells) are used in the reusable cells. None of the remaining 5 cells, loaded with standards, have any concern related to clay-mineral diffraction so Kapton^® is the norm for standard cells, but one of the standards (amphibole), useful for both XRD and EDH analysis, is loaded into a Mylar^®-windowed cell to provide access to at least one standard with this window design throughout the duration of the mission.

Once data from an observation are sent to ground and accepted, the analyzed material is emptied from the cell and that cell is ready to be reused. CheMin does not have the capability to store previously analyzed samples for later re-analysis once the wheel has been moved to receive another sample or to analyze one of the standards.

CheMin uses a 600×1182 pixel E2V CCD-224 frame transfer imager operated with a 600×582 pixel data collection area (Fig. 5). Once the 600×582 pixel active area is exposed to X-ray photons for a brief period (5–30 seconds), it can be rapidly transferred into a 600×600 pixel shielded area (the “transfer frame”). This allows data collection to take place continuously without the use of an X-ray shutter or electronic beam blanking. The pixels in the active portion of the array are 40×40 micrometers square, and the active region of deep depleted silicon is 50 μm thick. The front surface passivation layer is thinned over a substantial fraction of the active pixel area. This imager is a modern version of the E2V CCD-22 that was specially built for an X-ray astronomy application (Kraft et al. 1994). The large size of the individual pixels (relative to those present in conventional CCD imagers) causes a greater percentage of X-ray photons to deposit their charge inside a single pixel rather than splitting the charge between pixels. The enhanced deep depletion zone results in improved charge collection efficiency for high energy X-rays. The thin passivation layer makes the CCD sensitive to relatively low-energy X-rays (down to Si and in some samples, down to Al). The frame transfer region of the CCD (the portion of the CCD in Fig. 5 that is covered by a metal bar) has much smaller pixels than the active region because they are only meant to hold the charge associated with an X-ray photon, not to absorb it. An individual pixel in the array can hold hundreds of thousands of electrons, many more than the few thousand electrons deposited by a single X-ray photon. The active portion of the CD has larger pixels because the charge cloud of electrons that is deposited from an absorbed photon can be tens of microns in size.

Open image in new window

In order to keep the CCD from being exposed to photons in the visible energy range (from X-ray induced optical fluorescence from the sample) during analysis, a 150 nm Al film supported on a 200 nm polyimide film is suspended on a frame placed in front of the detector. The detector itself is cooled by a cryocooler to approximately 48 ^∘C below the temperature of the Rover Avionics Mounting Platform (RAMP) that is used as a heat sink by the cryocooler. The RAMP is expected to be between 0 ^∘C and +26 ^∘C during the mission, yielding CCD temperatures of −48 ^∘C to −22 ^∘C. By cooling, dark current in the CCD is reduced.

A number of interactions occur between the sample and the X-ray beam to produce photon fluxes detected by the CCD. The two interactions of importance to the CheMin experiment are elastic interactions in which primary beam photons are absorbed and re-emitted by sample atoms (i.e., diffraction events), and inelastic interactions in which sample atom inner shell electrons are ejected from the nucleus, resulting in the emission of a characteristic photon from the sample (i.e., fluorescence events). In X-ray diffraction, elastic interactions of primary beam photons with sample atoms constructively or destructively interfere, resulting in maxima in discrete angular directions (“Laue cones” or “Debye rings”) representative of the crystal structure of the sample. In fluorescence, sample-generated photons are emitted uniformly in all directions and the energies of individual photons can be used to identify the elements present in the sample.

X-ray diffraction analysis consists of the measurement of the directions and intensities at which crystalline matter diffracts X-rays. Placing an individual crystal in fixed orientation in a monochromatic X-ray beam will at most lead to a single diffracted beam, and most likely no diffraction at all. To accurately identify a crystalline phase, it must be exposed to the X-ray beam in all orientations to record all possible diffracted beams within the angular range covered. Powder XRD achieves this condition by using powdered materials—or solid polycrystalline materials—to create a sample that offers all possible crystalline orientations within the analytical volume. In laboratory powder-XRD instruments, fine-grained samples (ideally <10 μm grainsize) offer a very large number of crystallites in random orientations in the analytical volume. Miniaturized XRD instruments have even more stringent grain-size constraints because their analytical volume is scaled down. In an instrument like CheMin, an ideal sample would have a submicron grain size, difficult to achieve by grinding. When coarser than ideal powders must be analyzed, a means of increasing crystalline orientation statistics is necessary. A common method used in laboratories is to spin the sample in a thin glass capillary. For CheMin, a new approach was developed based on granular convection of the sample in vibrated cells. The analytical volume represents a small fraction of the sample cell volume (∼1/6000) but the internal flow of material through granular convection ensures that the entire sample is analyzed in random orientation over time (Fig. 6). Granular convection is obtained by vibrating the sample holder to fluidize the powder. Gravity combined with the interaction of the powder with the oscillating cell walls leads to convective flow. In addition to allowing quality XRD to be obtained with large grain sizes (up to 150 μm), the vibrations facilitate sample insertion and removal. Conventional methods of XRD sample handling would have required CheMin to be fitted with a grinder, a cell filling mechanism and possibly a cell spinner, increasing mechanical complexity, cost and risk. The benefits of the convective cells in producing XRD data suitable for quantitative analysis have been demonstrated in a number of CheMin precursor and laboratory analog instruments.

Open image in new window

A convective sample holder must ensure an intense vibration in the cell, thousands of m s⁻² (hundreds of g) being typically required to obtain reliable convection. This is achieved by using a balanced mechanical resonator similar to a tuning fork. A sample cell is integral to each arm of the tuning fork, and both cells are vibrated at the same time. The balanced design ensures minimal transmission of the vibration to the structure of the instrument; the energy is kept in the resonator. The vibration is induced by a piezoelectric actuator placed at the base of the tuning fork and driven at the resonance frequency of the assembly. The intensity of vibration is pulsed at about 1 Hz. A short period of intense shaking ensures that convection is initiated and a period of lower intensity maintains the motion while preventing the bed of powder from expanding and losing bulk density which would be detrimental to the XRD measurement.

Granular convection is a well-known phenomenon that finds few practical applications. It has been studied principally with reference to large beds of mm-sized particles, but little was known about the physics of granular convection in the micrometer range of the CheMin samples, or in the thin bed geometry of the CheMin sample cell. Granular flow in CheMin cells was modeled by Dr. C. Campbell (USC) using discrete particle computer simulations, and studied empirically by Particle Image Velocimetry (PIV). In PIV, contrast features (not necessarily individual grains—in fact the grains can be below the resolution of the observations) are identified and tracked from frame to frame in a sequence of captured video frames of particle motion. Vectors are drawn between the features to calculate granular flow velocities and bulk powder movement.

Both PIV and discrete particle computer simulations were carried out to better understand granular flow phenomena and predict the operation of the system at Mars surface conditions (lower gravity and lower pressure). The study of XRD data quality versus convection speed demonstrated that the velocity of convection has no real influence on quality of diffraction as long as the bed of powder is kept moving constantly. A pause in the motion will cause large grains to stay in a diffraction condition for an extended period of time, resulting in bright diffraction spots on the detector that anomalously increase the relative intensity of the corresponding XRD peaks. The main requirement for granular convection therefore, is that is must be robust.

Using PIV measurements, the velocity of granular flow was evaluated as a function of shaking conditions. It was found that above a specific vibration amplitude threshold, granular velocities are linear with amplitude for a given frequency (Fig. 7-left). The onset threshold for granular convection is reduced at higher frequencies (Fig. 7-right).

Open image in new window

The granular flow model was used to evaluate the effect of Mars gravity. Simulations run under Earth (E) and Mars (M) gravity (Fig. 8) show that at slightly higher velocities, flow patterns at Mars gravity are equivalent to those at Earth gravity. This is at odds with what is known about conventional granular convection in large beds of particles, for which the vibration required for a given velocity scales linearly with gravity. This difference in behavior is due to the predominance of grain-wall interactions in the thin cells of the CheMin instrument, as opposed to the predominance of inertial effects in thick beds.

Open image in new window

The effect of gravity was also studied experimentally by making PIV measurements of granular convection in a vibrated cell aboard a Piper Cherokee aircraft while performing a series of mini parabolic flights over the Pacific Ocean. Periods of 3–4 s at Mars gravity could be obtained. Slightly faster convection was observed under Mars gravity conditions. It was shown that only a small reduction of the vibration amplitude is required to maintain the same convection velocity from Earth to Mars gravity. For example a 30 μm peak-to-peak vibration at Earth gravity is equivalent to 25–28 μm at Mars gravity. Practically speaking, therefore, granular convection in Earth gravity is roughly equivalent to granular convection in Mars gravity.

The effect of the reduced atmospheric pressure on Mars was studied with the model and tested in a chamber at Mars pressure under dry conditions. No substantial effect could be associated with reduced pressure, although the transition from non-dry Earth conditions to low-pressure Mars conditions substantially slowed the convection. The transient state is due to water adsorbed on grains that increases the grain-grain cohesion when the pressure in reduced. Once the adsorbed water is removed, dry samples showed convection similar to that observed at ambient conditions.

The effect of electrostatic charging on granular flow under dry Mars conditions was also evaluated. A dedicated study was conducted by both modeling and experimental measurement of granular convection in a vibrated cell placed in a Mars atmospheric chamber. PIV measurements were made and charge accumulation on the windows measured. It was observed that granular flow initially creates a voltage on the windows by tribocharging, but this charge does not accumulate or persist. Most likely the charge is transported away by grain flow. PIV measurements showed no influence of this charge on granular convection. Based on this research, convective flow in vibrated sample cells on Mars should be similar to that observed on Earth.

Vibration is known to cause grain segregation when grains vary in size, density, shape or some other feature. Grain segregation in the CheMin vibrated cells is a concern because the analysis is performed in a very small volume in the middle of the cell. Preferential settling of a phase in any region of the cell would affect the measurement of mineral composition. Segregation in vibrated beds has been observed in particular when vibration amplitude is reduced to the point the convection no longer happens or happens very slowly, finer or denser particles are allowed to settle in lower regions of the cell. With higher vibration levels, the convective flow mixes the materials inside the cell at higher speed than segregation occurs, resulting in a cancellation of the effects of segregation in most cases. Segregation is still observed in the presence of convection in some extreme cases such as prepared mixtures of coarse grained materials with substantially different density (example: quartz + corundum). In order to limit this bias, the vibration profile includes a periodical shaking at very high amplitude for periods of a few seconds. This intense shaking is referred to as “Chaos mode” for the fast unstructured motion resulting in the cell, and has the effect of mixing and homogenizing the material. A case of apparent segregation in a beryl-quartz calibration sample data set from the flight instrument is presented in Sect. 4.2.1, thought to be due to the lack of chaos mode pulses during this particular measurement.

Another possible bias in the measurement of mineral composition is agglomeration of particles in the cell. This is observed at the edges of the cell in particular when the sample contains fines. The agglomerates are wedged between the polymer windows where they can least flex. These agglomerates grow over time as additional particles are jammed into them. As an example, crushed sandstone is likely to show a preferential agglomeration of the fine-grained cement matrix phases at the edge of the window, resulting in an apparent enrichment of the coarse sand particles (quartz) in the middle of the cell where the analysis is performed. The chaos mode also proves effective at limiting this agglomeration as long as the intense shaking pulses are applied frequently to break agglomerates in an early state of formation. If agglomerates are allowed to grow and compact over tens of minutes or hours, their compactness and resulting cohesion can make them difficult to break up and disperse with the chaos mode.

While segregation and agglomeration are of concern, methods have been found to mitigate or cancel them, and such methods have been validated by accurate quantitative analyses obtained with CheMin derived XRD instruments used at NASA Ames Research Center, NASA Johnson Space Center, Jet Propulsion Laboratory, Los Alamos National Laboratory, and Indiana University (see for example, Bish et al. 2008; Blake et al. 2009; Treiman et al. 2010; Achilles et al. 2011; Taylor et al. 2012). A good metric to evaluate the presence or absence of either segregation or agglomeration is to monitor the measured composition as a function of time during an analysis. Selective segregation or agglomeration progressively increases the proportion of one or more specific phases during an analysis. In such cases, more accurate compositional data are obtained at the beginning of the analysis, assuming the cell has not been shaken for an extended period between loading and start of the analysis.

All parameters can be set manually to simulate a specific geometry. Up to 3 parameters can be set as variables, scanned within a grid to study a range of configurations, and any set of parameters can be used as variables for optimization with a genetic algorithm approach. The data products of CheminRay include a 2-D image representing the distribution of diffracted photons on the CCD and a 1-D diffractogram resulting from the circumferential integration of intensity around the diffraction rings. An automatic line fitting tool imbedded in the application produces XRD peak intensities, peak shapes and peak positions which can be used for the optimization.

CheminRay is a virtual prototyping and testing tool. The advantage of ray tracing over other modeling approaches is that numerous parameters can be taken into account, yielding a numerical simulation with high fidelity to the real-world experiment. A drawback however, is that such simulations are time-consuming because as with real experiments the signal to noise ratio depends on the number of photons used in the model, hence the amount of computation time.

The starting point of the geometry optimization was the CheMin III instrument (a precursor to CheMin IV, the principal testbed of the CheMin flight instrument) available in the laboratory. The accuracy of CheminRay was first verified by comparing experimental data from CheMin III to simulated data of the CheMin III geometry as shown in Fig. 9. The CheMin III collimator design was then optimized using both a genetic algorithm and a grid scanning approach. Genetic algorithms apply techniques inspired by evolutionary biology to optimize a result by evolution from generation to generation of a population of potential solutions. In each generation, the fitness of the whole population is evaluated and individuals are selected and combined with a fitness-dependent probability to form a next generation of parameters. The formulation of “fitness factors” and “evolutionary rules” is critical in achieving an actual parameter optimization. The objective of the optimization of CheMin was to achieve the best resolution with the highest flux intensity. A number of fitness factor definitions were tested based on a mathematical combination of resolution and intensity, but all definitions proved to be too restrictive, leading to a particular throughput / resolution compromise. The genetic algorithm was modified to maximize the throughput of classes of XRD resolution as defined as the maximum FWHM observed in the XRD pattern. Computed populations of configurations were then reported in throughput versus resolution. After selecting a target resolution, the optimum configuration (in this case, a refinement of pinhole diameter and distance to the source) was obtained from the best configurations in the corresponding resolution class. Studies utilizing a grid scan led to similar results. The optimized geometry, which results in a resolution of 0.3^∘ 2θ was chosen for implementation in the CheMin IV instrument. A 20× increase in the XRD throughput (X-ray flux) was calculated compared to the CheMin III geometry, with a uniform 2θ resolution along the full range of 2θ (e.g., see Fig. 10).

Open image in new window

Open image in new window

Similar optimization approaches with more parameters were taken to explore the potential of different instrument parameters for the flight instrument. In particular, optimized geometries were proposed based on a dual X-ray tube layout as specified in the early stage of the flight instrument development. After the second X-ray source was descoped, an instrument geometry was proposed based on a slight evolution of the CheMin IV design, having a single X-ray tube, and sample and CCD detector both perpendicular to the beam. CheminRay was used to study the effect of parameters such as pinhole diameter or sample thickness on resolution, intensity, peak asymmetry and peak shift. Ray tracing results were backed by experimental verification of selected geometries using a dedicated breadboard instrument based on commercial grade components.

Once the geometry of the system was established, a second model was developed based on a mathematical description of the diffracted intensity at the CCD taking into account all critical parameters of the geometry. Unlike the ray tracing model which could virtually explore all possible configurations but required extensive computation time, this model limits its scope to the chosen geometry through a number of approximations, but requires far less computation time. The output of the mathematical model was verified with ray tracing results as well as experimental data from breadboard instruments. This model was a critical engineering tool for the design of the flight instrument, allowing detailed investigation of the effect of small variations of design parameters on the resulting patterns. The model was used primarily to link engineering requirements to MSL level II science objectives. Small but significant adjustments were made to the geometry using this tool.

Line profile fitting is the basic approach to obtain phase contents and structure information from X-ray powder diffraction data. Typically, two methods are used for line profile fitting: Constrained or unconstrained analytical profile fitting, and the fundamental parameters method.

In the first method, empirical instrument profile functions (mathematical functions) are used to simulate the observed diffraction pattern. In some cases, these mathematical functions are convolved with other empirical descriptions of instrument- and sample-related contributions to generate a simulated diffraction pattern. The disadvantage of this method is that refinable variables relating to the instrument profile functions have no clear physical meaning; they are simply empirical mathematical functions. Profile parameters resulting from a refinement are generally not sensible, even though refinements converge well, giving acceptable residual factors. In such cases, it is often common to apply mathematical constraints to the refinement to obtain reasonable results.

In the case of fundamental parameter descriptions of profiles, all of the geometrical parameters of the entire instrument are mathematically described and are convolved together. In this case, results of refinement have definite physical meanings. By convolving the individual instrument parameter functions together, an instrument profile can be generated, and if these functions accurately describe the instrument, the simulated instrument profile can be very accurate (as determined through the use of profile standards such as NIST SRM 660a LaB₆). This instrument profile can then be convolved with sample-related profiles to simulate the experimental diffraction pattern. Generally the individual instrument contributions to the observed profile are well known, for example, the diffraction geometry, slits, radius, etc., employed in a Bragg-Brentano powder diffractometer. Some instrument-related parameters are not precisely known, and refinement of these within a limited range can ultimately provide instrument functions that have physical meaning, in contrast to the analytical profile functions used in the empirical line profile fitting method. As a result, structure and phase information resulting from a refinement are more realistic and accurate.

The geometry of the CheMin instrument is shown schematically in Fig. 1. X-rays are generated by a Co radiation source, they are shaped by a collimator, and they are ultimately diffracted by the sample. The instrument profile for this configuration can be simulated by convolving the radiation source spectrum of Co with a function for the beam size aberration, a function for the collimator, and a function accounting for sample thickness. The source emission profile is generated by convolving seven Co Kα lines and six Co Kβ lines with a Lorentzian function (Hölzer et al. 1997). The Kβ lines in the diffraction pattern originate from radiation leakage (e.g., Kβ photons that have lost charge (“split events”) and fall into the Kα energy window), and the intensities are thus allowed to be refinable, maintaining their relative intensities. The finite X-ray beam size effect is approximated by a Gaussian function rather than an impulse function (Young 1993), and the collimator is described by a circular function.

Absorption Correction for the Normal-Beam-Transmission Diffraction Experiment

The diffraction geometry for the CheMin instrument is a normal-beam-transmission geometry. As shown below, diffraction intensities vary systematically with diffraction angle 2θ for this instrument geometry resulting from sample absorption (which is not the case in Bragg-Brentano geometry). Thus an absorption correction must be applied to obtain accurate diffraction intensities for Rietveld refinement, in order to provide reasonable phase contents and structural details. With reference to Fig. 11, I₀ is the diffraction intensity of a unit volume sample at angle 2θ under the condition of zero absorption, S is the cross-sectional area of the direct beam at position x in a sample of thickness t,x and d are the path lengths of the X-ray in the sample before and after diffraction, and μ is the linear absorption coefficient of the sample. The diffracted intensity from the volume element Sdx with diffraction angle, 2θ, is:

$$ dI=I_0 e^{-\mu(x+d)}Sdx=I_0 Se^{-\mu[x+(t-x)\sec 2\vartheta]}dx $$

(1)

Open image in new window

The total diffracted intensity at diffraction angle 2θ is the integration of dI at the same angle over the range of sample thickness x∈[0,t].

Open image in new window

(2)

Relative intensities are used instead of absolute intensities in Rietveld refinement, in both the analytical profile fitting and fundamental parameter method, and therefore the intensities at diffraction angle 2θ can be normalized to the intensities at 2θ=0:

$$ I_{2\theta=0}=I_0 \mathit{St} e^{-\mu t} $$

(3)

The normalization factor for the intensity correction is

$$ f_{2\vartheta} =\frac{I_{2\vartheta}}{I_{2\vartheta=0}}=\frac{e^{\mu t (1-\sec 2 \vartheta)} -1}{\mu t (1-\sec 2 \vartheta)} $$

(4)

There are similarities between the experiments as they were conducted on the Inel XRD for V&V, and as they will be performed on the CheMin DM or FM instruments. These similarities principally involve the sample cell and the environment in which the sample is analyzed. Of all of the CheMin testbed instruments available (with the exception of the DM), only the Inel instrument can collect diffraction patterns utilizing an MSL cell in a 7 torr atmosphere. The principal use of the Inel machine is to study, evaluate and understand grain motion during diffraction. The following experiments were conducted in the Inel testbed during V&V.

Several Terra^® instruments (a commercial spinoff instrument of CheMin, manufactured by inXitu, Inc., Campbell, CA) are in use at Ames Research Center, Jet Propulsion Laboratory, Los Alamos National Laboratory, Johnson Space Center and Goddard Space Flight Center. They are battery-operated, field-portable instruments designed with enough fidelity to the CheMin FM to allow practical evaluation of how the CheMin design functions in field situations and for the development of XRD libraries. The Terra^® instrument at GSFC is being used for direct comparisons of mineral and rock analyses between SAM and CheMin testbeds. The Terra instrument at JSC is likewise being used for direct comparisons of mineral and rock analyses between CheMin and the ChemCam testbed at LANL.

In addition to the Prototype 1 assembly, a separate funnel assembly was used to evaluate funnel functions. The funnel assembly was used to examine grain movement and amount of remnant contamination with a variety of samples. The stand-alone funnel was tested at angles up to the maximum tilt design for CheMin FM operation (up to 20^∘ with the rover parked). The funnel assembly tilt tests were done in the same Mars chamber used to test samples with Prototype 1.

The flight model (FM) has elemental and XRD standards in five permanently loaded sample cells (discussed below in Sect. 3.3). Prior to integration with the rover, these standards were analyzed in a simulated Mars atmosphere at a range of RAMP temperatures to obtain baseline function data from the FM, for (1) defining initial pre-mission functionality and (2) cross-calibration with identical samples that are loaded in the demonstration model (DM). Results of FM analyses of the permanently loaded standards are shown in Sect. 4.

The demonstration model (DM) was built from components identical or equivalent to those used in constructing the FM. Identical standards to the FM are loaded in five permanent cells and are used to evaluate any changes in functionality of the FM that might be related to ATLO, launch, flight, and landed operations. The DM will be cross-calibrated to the FM using these common standards. DM testing prior to and during the mission with a variety of samples is critical to interpreting FM data, because no end-to-end testing was performed in which samples were loaded through the funnel and analyzed in one of the 27 open sample cells of the FM (doing so would have exposed the inside of the instrument to contamination from sample powder with unknown consequences during launch and spaceflight).

The DM will be kept on-site at JPL for purposes of (1) continuing evaluation of DM performance relative to results returned by the FM; (2) prototyping possible changes in function (analytical sequence, piezovibration operation, temperature cycling, etc.) before applying such changes to operation of the FM on Mars; and (3) analyzing selected known mineral and rock samples in open cells, which could not be performed in the FM.

Detection limits for individual phases as well as the precision and accuracy of mineralogic analysis are influenced by geometrical factors. Because the CCD is off-center and has a finite size, diffraction rings are not fully captured by the CCD. The proportion of each ring captured by the CCD is a function of 2θ. At high 2θ values, less of the ring is captured. At low 2θ, the effective radius of the diffraction ring is so small that relatively few pixels define the ring. A diffracted peak at very high or very low 2θ will consequently have poorer statistics than one at moderate 2θ. Of equal importance, the CCD has a specific Charge Collection Efficiency (CCE) for Co Kα and Co Kβ photons (i.e., as a function of photon energy, only a fraction of the photons that strike the CCD will be absorbed by the silicon), as well as a characteristic energy-dependent and device-dependent ratio of single events vs. split events (photons that are of higher energy and are absorbed deeper in the silicon have a tendency to form an electron charge cloud that diffuses across pixel boundaries). Each of these factors will influence the statistics of the final diffraction pattern.

Crystalline standards having diffraction maxima across the range of observed 2θ are sufficient to determine and calibrate 2θ, 2θ range, and 2θ resolution values for the DM and the FM. Of importance to this calibration (and to all diffraction data from CheMin) is the random orientation of crystallites presented to the beam during an analysis. For these measurements, a standardized powder size range is used (45–90 μm grain size; this same size range is used in loading all of the standards in the FM and DM) as well as a standard protocol for shaking the sample. Under normal circumstances, the grain size distribution of the sample will be a function of the SA/SPaH sampling system as well as the nature of the sample.

An important parameter in quantitative XRD is the instrumental broadening function (described in Sect. 1.7.3). In a sample of finite thickness, diffraction can take place anywhere along the beam path through the sample. Once diffracted, photons can be reabsorbed or rediffracted along their path inside the sample. In weakly absorbing samples, diffracted photons originating from near the front surface of the sample (X-ray source side) will contribute proportionally more to the pattern than in strongly absorbing samples. This results in an asymmetry in the individual diffraction peaks that is a function of the linear absorption coefficient of the material for Co Kα and Co Kβ photons—and the thickness of the sample.

Absorption by the sample of transmitted Co Kα and Co Kβ photons will vary as a function of composition. In addition, the mass thickness of the sample will vary depending on the proportion of solid material and void space present in the beam as the sample is shaken and different components of the powder pass through the beam. Granular standards are used to ensure that these materials are as much like the unknowns as possible, to account for such effects.

A mixture of two minerals with several closely spaced diffraction maxima is used to calibrate 2θ resolution at low, medium, and high ranges of 2θ. The minerals selected were beryl (a synthetic emerald donated by Chatham Created Gems, Inc., San Francisco, CA) and quartz. Grain size distribution is controlled (45–90 μm) to minimize grain size effects and cell loading is standardized to control particle mass thickness along the pathways of diffracted photons.

Known standard minerals are used to calibrate both Co Kα and Co Kβ diffraction patterns for peak locations and relative peak heights. Minerals chosen were amphibole, beryl, quartz, and arcanite. Grain size distribution is controlled (45–90 μm) to minimize grain size effects and cell loading is standardized to control particle mass thickness along the pathways of diffracted photons.

For mineral identification, the search/match utilities in commercial XRD analysis programs such as Jade^® (MDI, Inc., Pleasanton, CA) and Xpowder^® (Martin 2004) will be used. During the mission, “Payload Downlink Leads” will analyze CheMin data in a tactical timeframe (minutes) to produce “quicklook” data products for use in planning the next Sol’s activities (these qualitative analyses will use search-match programs such as Jade^®). Following mineral identification, quantification will be accomplished using one or more of several established quantitative analysis methods on a non-tactical timeframe (weeks). For samples in which constituent minerals do not have known crystal structures, or minerals and components without well-defined structures (e.g., glasses and many clay minerals) a program such as FULLPAT will be used. In FULLPAT analyses, individual library standard patterns for all phases believed to be present in the sample are simultaneously fit to the observed pattern using a least-squares refinement. These library patterns can be calculated from crystal structures, simulated from ICDD data, or directly measured on the DM or testbed instruments. A version of FULLPAT tailored specifically to CheMin instrument parameters is being developed by co-I Steve Chipera. Several hundred mineral, rock and soil samples, including phyllosilicates, carbonates and evaporites, are being measured in the CheMin IV instrument at Ames Research Center, for use in FULLPAT analysis. A subset of this sample suite will be analyzed on the CheMin DM instrument at JPL. For samples composed of minerals with well-known crystal structures, full-pattern fitting and Rietveld refinement will be used, in conjunction with Jade^® or TOPAS^® software to determine mineralogical compositions. For Rietveld refinements, patterns are calculated from first principles for all the phases in the sample, using known crystal structures. The calculated patterns are used to simulate the observed pattern and the quantitative abundances are obtained directly from scale factors. There are advantages to both methods. FULLPAT can be used for phases that are not fully characterized or whose three-dimensional structures are poorly constrained, such as many clay minerals and X-ray amorphous materials. The Rietveld method is potentially more accurate but requires a full description of the crystal structures.

Several potential standards were examined in laboratory diffractometers, testbeds, and by other methods (e.g., thermogravimetric analysis; XRF analysis; electron microprobe analysis) to determine suitability as standard materials. Five were selected for use (Table 11). Two cells in the FM and DM carry quartz:beryl mixtures (3:97 and 12:88) that are required for calibration and verification and validation of the FM. Three other cells are loaded with standards chosen for varied fluorescence with diverse chemical composition, namely amphibole (Fe-containing, to measure Fe Kα and Co Kα separation in energy-dispersive histograms), arcanite (a sulfate mineral), and a synthetic ceramic containing a wide range of elements (to understand FWHM of individual elemental peaks as a function of energy and CCD temperature in the EDH, as well as split-pixel effects).

Table 11

Flight model (FM) and demonstration model (DM) standards: composition (weight %), linear absorption coefficients for Co Kα, reference intensity ratio, density, and mohs hardness

Open image in new window

*Light-element components in red italics are not be detectable by CheMin and are shown for completeness only

**RIR is the reference intensity ratio: the ratio of the strongest peak of the phase of interest to that of the (113) corundum peak in a 1:1 mixture (50-50 by weight)

Table 11 lists the five standards that were chosen. Column 1 lists an amphibole standard (from Gore Mountain, New York) with both an acceptable range of diffraction maxima and varied chemical composition that includes high Fe content. Columns 2–3 list the two standards selected specifically for strong diffraction over a range of 2θ (beryl:quartz). Columns 4–5 list a high-fluorescence S-rich composition (arcanite), and the chemically diverse doped ceramic. Table 11 also lists the hardness and density for each of the standards selected. Density is an important factor in determining which minerals can be mixed together in a single cell (e.g., beryl and quartz); minerals of comparable density are less likely to exhibit grain segregation during piezovibration of the sample cell. Hardness is also a factor if different minerals are mixed together, as the harder minerals will tend to abrade softer minerals.

The amphibole has several closely-spaced diffraction maxima that provide a means of evaluating 2θ resolution and peak-profile parameters from low 2θ (∼12^∘) to 48^∘ 2θ. This standard also has a relatively complex composition (magnesiohastingsite or pargasite, depending on ferrous/ferric ratio), useful for evaluation of CCD EDH performance; composition determined by INAA and electron microprobe is shown in Table 11. It has a significant Fe content that can be monitored to evaluate any overlap of Fe Kα into the energy range selected for analysis of the diffracted primary Co Kα used for X-ray diffraction. There are no known stability issues with use of amphibole as a diffraction standard.

The beryl used in the DM and FM is a synthetic emerald supplied by Thomas Chatham of Chatham Created Gems, Inc. The sample from Chatham used in the DM and FM contains 0.73 weight percent Cr₂O₃ as determined by INAA; other constituents (Be, Al, Si) will not be detected or are poorly determined by EDH with CheMin. Beryl has three well-defined and separated peaks of high intensity (I>60) across a 2θ range from 13^∘ to 36^∘ in Co Kα radiation. It also has a high reference intensity ratio (RIR of 2.1), providing a strong signal after relatively short exposures. There are no known stability issues with the use of beryl as a diffraction standard. Beryl has diffraction maxima that are relatively simple and widely spaced, providing a capability for evaluation of peak profiles. The small amount (3 % and 12 %) of quartz mixed with this sample for V&V purposes will not interfere with XRD calibration based solely on the beryl component.

Arcanite in the DM and FM is synthetic, prepared by crystal growth from saturated solutions of reagent K₂SO₄. The stoichiometric composition is listed in Table 11. The arcanite has high K and S content that will produce fluorescence at 3.31 and 2.31 keV, respectively, creating fluorescence background in the diffraction pattern in a manner comparable to some of the sedimentary samples likely to be examined by the MSL rover at Gale Crater. In addition, arcanite has diffraction maxima that will provide a means of evaluating 2θ resolution and peak-profile parameters from ∼25^∘ to 48^∘ 2θ. It has the highest sulfate content among the standards. Arcanite is rare among sulfate salts in that it does not readily hydrate; as such it provides an exceptionally stable sulfate calibration standard. It is also very stable on heating.

The doped ceramic used in the DM and FM is synthesized from 41 % of the Clay Minerals Society nontronite NAu-2 (pre-fired to 1000 ^∘C to avoid vessiculation during later firing) mixed with smaller portions of calcium phosphate, anhydrite, potassium bromide, rutile, zinc sulfate, Mn-sulfate, rubidium chloride, strontium chloride, chromium oxide, nickel chloride, and 9.5 % Li-tetraborate flux. This mix was ball-milled, pressed, and fired in a porcelain crucible to 800 ^∘C (firing temperature is kept low to minimize sulfur loss). The ceramic produced is a fine-grained mixture of complex synthetic phases with a small amount of borosilicate glassy matrix. This material was crushed and sieved to 45–90 μm to provide a polycrystalline sample containing most of the elements of interest to the MSL mission that are also within the detection range (13<Z<42) of the CheMin CCD. The analysis listed in Table 11 is calculated for the as-fired ceramic before crushing and sieving.

The candidate standards must be able to survive thermal and vacuum testing, decontamination bakeout, and several months at space vacuum, followed by two Earth years (one Mars year) in the MSL rover body on Mars. Of these conditions, decontamination bakeout at 80 ^∘C and 10⁻⁵ torr is most severe. Unstable minerals, especially hydrous phases that readily exchange water, are not suitable. Quartz and beryl contain no water, hydroxyl, or other potentially volatile constituents and will remain stable. To assess stability of the arcanite, ceramic and hornblende, thermogravimetric data were collected to 800 ^∘C with dry N₂ flush (overheating to 10× the expected maximum temperature compensates for lack of vacuum conditions). Of the standards to be flown, arcanite showed no weight loss. Dehydroxylation of amphibole occurs at high enough temperature (>400 ^∘C) that stability should not be an issue. A small weight loss begins at ∼400–500 °C in the ceramic and it too will be stable throughout the mission.

Cross-calibration of the CheMin DM and FM is based on the five standards carried in and common to both. Extended calibration using the DM and other instruments, principally CheMin IV, is planned and results will be reported later. The CheMin DM and the CheMin IV instruments will be used to analyze blind samples provided to the CheMin science team by the MSL Project. The Project will make available several different samples for a round-robin exercise. XRD results on the blind samples using the CheMin DM and CheMin IV instruments will be compared to the conventional analyses as a check on CheMin calibration and performance.

In addition to the blind samples provided by the MSL project, the CheMin science team is analyzing its own suite of 140 blind samples provided by CheMin Co-I Dick Morris of Johnson Space Center as a round robin exercise. Many of these samples are also being analyzed by other MSL instrument testbeds such as ChemCam and SAM. Once the samples are analyzed using the CheMin IV testbed instruments at JPL, NASA/ARC and NASA/JSC, the compiled analyses will be compared with results obtained by XRD in the laboratory of CheMin Co-I David Bish of Indiana University. Selected samples from this suite will be analyzed in the DM as deemed appropriate.

Instrument XRD and EDH performance checks were conducted on the FM during ThermoVac testing and will be conducted during the deployment of the FM on Mars, to evaluate instrument degradation and instrument drift. Instrument performance checks will be conducted using the same algorithms developed for routine data analysis (Sect. 1.9). Performance checks will be conducted in concert with routine calibration, approximately once every 40 sols, or as needed should data downlink indicate off-normal operation.

In this section, analyses of a suite of ultramafic rocks similar to those identified on Mars (Poulet et al. 2009; Ehlmann et al. 2009) are shown. We evaluated the ability of the Terra instrument to identify and quantify minor minerals in ultramafic mantle xenoliths and we evaluated the accuracy of analyses by comparing XRD results with thin section point counting measurements (optical petrography and electron microprobe analysis). Mantle xenoliths were collected on the 2008 AMASE (Astrobiology Mars Analog Svalbard Expedition) expedition to Svalbard (Steele et al. 2008), from basalt outcrops on the Sverrefjell volcano (Skjelkvale et al. 1989). Each xenolith was split into several fragments (a few gm each) for thin sectioning, XRD, and petrographic analyses.

Mineral identifications were first made by optical microscopy. Mineral proportions were derived from optical images of the whole thin section area, manually annotated for mineral species and measured by area in an image-processing code (NIH Image; http://rsbweb.nih.gov/ij/). Weight percent was calculated from area percent by multiplying area percent by assumed phase densities. We believe that the proportion of total pyroxenes is accurate, but it was difficult in some samples to distinguish orthopyroxene from clinopyroxene. Mineral compositions were determined by wavelength-dispersive electron microprobe analysis (EMP) using well-characterized natural and synthetic standards (Treiman et al. 2010).