X-RAY powder diffraction (XRPD) analysis has been

Transcription

1 Published online March 1, 2007 Influence of X-Ray Diffraction Sample Preparation on Quantitative Mineralogy: Implications for Chromate Waste Treatment Dimitris Dermatas, Maria Chrysochoou,* Sarra Pardali, and Dennis G. Grubb ABSTRACT Powders of chromite ore processing residue (COPR) were mineralogically evaluated using quantitative X-ray powder diffraction (XRPD) to illustrate the impacts of sample preparation procedures. Chromite ore processing residue is strongly alkaline, reactive, contains minerals of varying hardness and absorption coefficients, and exhibits significant amorphicity. This poses a challenge to produce powders for XRPD analysis that are sufficiently fine and of uniform particle size while avoiding mineral reactions and overgrinding effects. Dry, hand pulverization to different grain sizes, and wet, mechanical pulverization (micromilling) using four milling liquids (cyclohexane, isopropanol, ethanol, and water), and variable milling durations (up to 15 min) were evaluated. Micromilling with a light, nonpolar, highly evaporative liquid such as cyclohexane with a milling time of 5 min mitigated systematic errors such as microabsorption and preferred orientation as it produced finer and more uniform particle size distributions than the hand-pulverized powders, while simultaneously affording the least time for sample preparation. Conversely, the use of water as milling liquid resulted in extensive hydration reactions during sample preparation, causing mischaracterization and significant underestimation of its reactive brownmillerite content, which can complicate the remediation design process for COPR. Hand pulverization emerged as a necessary complement to quantify Cr(VI)-containing, softer minerals destroyed during mechanical milling, the quantification of which has also important implications for COPR treatment design. The findings of this study may be applicable in a variety of geochemically complicated and reactive environmental media (metal-contaminated soils, stabilized/solidified media, inorganic waste), and points to the importance of the sample preparation method to obtain reliable quantitative XRPD results. W.M. Keck Geoenvironmental Lab., Castle Point on Hudson, Stevens Inst. of Technology, Hoboken, NJ 07030, USA. Received 1 June *Corresponding author (mchrysoc@stevens.edu). Published in J. Environ. Qual. 36: (2007). Technical Reports: Heavy Metals in the Environment doi: /jeq ª ASA, CSSA, SSSA 677 S. Segoe Rd., Madison, WI USA 487 X-RAY powder diffraction (XRPD) analysis has been used in many comprehensive studies over the years for the qualification and quantification of crystalline and amorphous phases in solid matrices for a wide variety of purposes and decision making. The quantitative techniques, however, are relatively new to the environmental chemistry and remediation fields. While the environmental regulatory framework relies on total concentration or leaching behavior of contaminants to delineate environmental contamination, XRPD has the potential to identify the in situ contaminant speciation, not achievable via other conventional chemical analyses, and thus it constitutes a critical analytical step toward the reliable prediction of contaminant mobility and geochemistry. X-ray powder diffraction technology and quantitative procedures have matured to the point where the information provided can now be leveraged to provide insight into the speciation of metals in contaminated environmental systems when used in combination with other environmental investigative techniques. This has a twofold effect: if we know where the metals are we can: (i) understand their true environmental hazard; and (ii) use knowledge of metals speciation/mineralogy to optimize management systems or remedial solutions. Qualitative XRPD has been widely employed in the environmental literature to investigate metal speciation, as it is a nondestructive method that is widely available in research laboratories. While quantification of crystalline phases identified in XRPD patterns by the Rietveld method (Rietveld, 1969) is routinely used in disciplines such as cement research, it has known very limited application in the environmental field. Hillier et al. (2001) applied Rietveld to quantify the amount of cerussite (PbCO 3 ) and to investigate the speciation and mobility of Pb in contaminated sediments. Hillier et al. (2003) and Chrysochoou and Dermatas (2006) used the same technique to quantify the mineralogy of chromite ore processing residue (COPR) and delineate the speciation of hexavalent chromium (Cr(VI)). In principle, when a metal precipitates as a crystalline compound that can be identified by XRPD, its quantification can provide insight into the bound quantity and the long-term stability. Also, the comparison of quantitative XRPD data with the total metal contents measured by wet chemical analyses provides an estimation of the amount of metal present in other forms, nondetectable by XRPD. The application of quantitative XRPD analysis in the environmental field may provide useful information on metal and contaminant speciation, and enable delineation of the geochemical properties of the media evaluated. In the case of certain types of COPR, derived from the high-lime process, the geochemical stability of the material is directly related to its brownmillerite (Ca 2 FeAlO 5 ) content, the predominant mineral formed during the roasting process from which COPR is derived. In short, lime (CaO) was added to chromite ore ((Fe,Mg)(Cr,Al) 2 O 4 ) at 1200jC; Cr was removed as soluble chromate (CrO 4 22 ), while the remaining metals were complexed with Ca and removed as solid waste (COPR). Chromite ore processing residue was widely used as fill in the marshlands found in various U.S. states, including New Jersey and Maryland. A detailed description of the history of COPR and the issues associated with its deposition in the U.S. can be found in Chrysochoou and Dermatas (2006) and Dermatas et al. Abbreviations: COPR, chromite ore processing residue; XRPD, X-ray powder diffraction; RQA, Rietveld quantification analysis; MA, microabsorption; PO, preferred orientation; SEM, scanning electron microscopy.

2 488 J. ENVIRON. QUAL., VOL. 36, MARCH APRIL 2007 (2006). Apart from its Cr(VI) content, which renders COPR a hazardous waste, it also exhibits unfavorable geotechnical behavior. Its delayed volumetric expansion promoted concrete failure and demolition of buildings at a deposition site in Hudson County, New Jersey. Dermatas et al. (2006) used quantitative XRPD to demonstrate that the rapid transformation of thermodynamically unstable brownmillerite to ettringite (Ca 6 Al 2 (SO 4 ) 3 (OH) 12 I26H 2 O) was responsible for pronounced heave during a field treatability study involving ferrous sulfate. The amount of brownmillerite therefore determined the reactive potential of COPR in the presence of sulfate. Heaving phenomena were also observed in untreated COPR deposited at the same site, and was also linked to hydration reactions of brownmillerite (Moon et al., 2006). Quantitative XRPD emerged as the only available tool to monitor the brownmillerite content in untreated and treated COPR samples and evaluate the reactive potential of COPR. There is little or no mention in the environmental literature with regard to XRPD sample preparation. A cursory review of studies employing qualitative XRPD to delineate metal speciation showed that while the Materials and Methods section typically provided a description of the characteristics of the X-ray equipment and the scanning parameters, there was no detailed description of the XRPD sample preparation methods. This is partly a result of the notion that the sample preparation method does not influence the results of qualitative analysis, as well as a default approach to prepare samples by simple pulverization. Apart from the quantitative studies mentioned above (Hillier et al., 2001, 2003; Chrysochoou and Dermatas, 2006), only one reference in the environmental literature was identified that specifically addressed the importance of XRD sample preparation techniques in qualitative analysis of zerovalent iron materials for permeable reactive barriers (Phillips et al., 2003). The authors concluded that the choice of drying method (acetone- vs. air-dried) and sample fractionation significantly affected the phase assemblage and the conclusions drawn from the mineralogical analyses (Phillips et al., 2003). Sample preparation and particle size are certainly decisive when quantitative analyses of complex, reactive, solid waste matrices are attempted (Bish and Post, 1989). Von Dreele and Cline (1995) stated that the precision, or repeatability, of measurements of relative phase abundance via conventional XRPD equipment can be better than 1% when care is taken in the proper specimen preparation. According to Hillier (2000), no matter how sophisticated the process of data treatment, there can be no doubt that sample preparation is the first key step in quantitative analysis, not only in terms of reducing potential sources of error, but also in terms of its practical application. Moreover, the use of coarse powders in XRPD work may have implications, such as: (i) limited depth of penetration of the X-ray beam and loss of information on coated particles; (ii) strong preferred orientation (PO) of platy crystals and thus loss of resolution to identify minor phases; and (iii) strong microabsorption (MA) effects that can bias semiquantitative analyses based on relative intensities. Thus, errors introduced during sample preparation will most likely lead to mischaracterization. Accordingly, due to the increased use of XRPD to delineate metal speciation, mechanisms of transformation, and attenuation, mineralogical mischaracterization may have important regulatory consequences and may lead to unnecessary, inappropriate, and/or inefficient remediation designs and site management strategies. Qualitative and quantitative changes in specimen mineralogy occurring during sample preparation therefore emerge as a central issue where the overall accuracy of XRPD is concerned. The actual quantity of the specimen analyzed with XRPD is rather small (2 20 mg), depending on the beam penetration (Bish and Post, 1989). Its preparation requires the production of a powder, i.e., pulverization, unless the material tested is naturally fine-grained. Dry and wet, hand, and mechanical pulverization techniques have been employed in the literature to achieve finegrained specimens (Bish and Post, 1989). Jenkins et al. (1986) provide an overview of the general considerations for XRPD sample preparation methods. However, there are no explicit guidelines (such as ASTM or otherwise) for the preparation of the required powdered mass, or to evaluate the impacts of the sample preparation procedures (pulverization technique, milling liquid, duration) on the initial characteristics of the media tested, especially for reactive or chemically unstable waste matrices. Investigators are left to independently develop a sample-specific preparation procedure which is believed to least perturb the initial mineralogical characteristics of the analyzed media. While the materials science community is generally cognizant of XRPD sample preparation and analysis techniques (and their pitfalls), environmental professionals are less attuned to these challenges and their consequences. As such, the purpose of this study was to highlight the complexities associated with the quantitative XRPD analyses of reactive contaminated matrices, using COPR as the example. Being an essential part of sample preparation, pulverization techniques and particle size reduction may introduce several artifacts which impact the initial and/or interpreted mineralogy of the powdered material. Sample disturbance is a major concern with sensitive media such as COPR due to its high ph, meta-stable state, and susceptibility to reactions. In a sense, COPR is an excellent media with which to demonstrate the influence of the sample preparation procedures on the subsequently performed XRPD analyses and Rietveld quantification analysis (RQA). An abbreviated theoretical background has been included below to acquaint the inexperienced XRPD users with the key issues in XRPD sample preparation. BACKGROUND Factors affecting the quality of the analyzed powder specimen during its preparation include: (1) Powder reactivity (2) Particle hardness (3) Particle size distribution (4) Choice of internal standard

3 DERMATAS ET AL.: X-RAY SAMPLE PREPARATION OF CHROMATE WASTE 489 Powder Reactivity Phase transformations and reactions may occur during dry or wet grinding, especially when unstable and/or reactive materials are processed. Temperature increases observed during dry mechanical milling can induce reactions in the powder. Wet grinding may mitigate temperature effects, but at the same time initiate and influence reactions occurring in the powder when the milling liquid and its properties have not been carefully considered. Water, isopropanol, and cyclohexane are often recommended as potential milling liquids by the manufacturers of mechanical grinders (McCrone Micronizing Mill), the first two liquids being also the most widely used (Puledda and Paoletti, 1994; Hillier et al., 2003). Ethanol has been also utilized in the grinding of cementlike materials and fly ash (Winburn et al., 2000a). The choice of liquid affects the sample preparation time and the extent of reactions that may occur during the powder preparation process. The use of water can affect unstable and hydrating compounds, as is the case with cements and cementitious waste, including COPR, while the use of organic liquids affects crystalline organic compounds; but the latter case is not generally pervasive in environmental characterization and remediation projects. After wet mechanical grinding, time is required for the powdered solids to settle out of the slurry before decanting the excess liquid. Faster precipitation of particles occurs in lower viscosity liquids (by Stokes law) and nonpolar liquids (allowing gravitational forces to dominate electrostatic forces). The remaining slurry has to be desiccated to obtain a dry powder. Rapid evaporation of the milling liquid is therefore highly desirable to avoid hydration and carbonation reactions in the powder, especially when drying occurs under atmospheric conditions (exposure to CO 2 and H 2 O). The use of gloveboxes can prevent sample contact with the atmosphere, but significantly prolongs the drying time and adds unwanted complexity to the sample preparation process. Particle Hardness The ability to obtain uniform particle size distribution in an XRPD powder is related to the crystal hardness of each mineral. As such, softer minerals may produce relatively finer powders and may also be additionally ground by friction with the coarser particles, both of which skew the mineralogical composition of the powder with respect to gradation (fractionation). In other words, the finer fraction will primarily consist of softer minerals, while the coarser fraction of harder phases. Though short-sighted, it is a common practice during hand-pulverization to pulverize and sieve only the material necessary for the XRPD specimen and not the total quantity of the sample, since hand pulverization can be an arduous and time consuming process. Periodical sieving of the powdered sample, however, only removes material of sufficiently small size (mainly softer crystals). As such, if the entire sample is not completely processed, the produced powder may not be representative of the initial sample mineralogy and its proportions, i.e., it will be fractionated. Hand-pulverization may enhance such phenomena due to lack of adequate control on the applied grinding pressures. Wet milling with mechanical grinders significantly increases the efficiency of grinding by uniformly reducing the particle size of the hard minerals, while not overgrinding the softer ones (Bish and Post, 1989). The liquid slurry helps ensure that the sample does not compact into the corners of the milling jar where it escapes the grinding elements. Additionally, comparisons of dry and slurry mechanical grinding have shown that slurry-ground products always have the narrowest particle size ranges, and for comparable grinding times, slurry grinding produces a finer powder (Bowden and Tabor, 1950). Particle Size Distribution The particle size distribution of the powder is affected by insufficient grinding or overgrinding. Accurate XRD intensities generally require that the grain size be small (,10 mm) and preferably 1 mm (Jenkins et al., 1986; Bish and Post, 1989). The main challenges involved in properly analyzing coarser-grained powders (.10 mm) involve microabsorption and/or preferred orientation of specific crystals. Microabsorption is related to the variable ratio of scattered vs. absorbed X-rays of individual minerals in a mixture depending on their linear absorption coefficient (m)andparticlesize.xrpdquantitativeanalysisisbased on the assumption that the reflected intensity is only a function of the mineral properties (reference intensity ratio or RIR) and the amount of the mineral in the sample. However, the scattered or reflected intensity is also a function of the ability of the mineral to absorb X-rays, which is described by m. Microabsorption skews RQA when there is significant contrast in the m values between the different phases, and when the powder is coarse. Microabsorption may be relevant to various environmental projects, e.g., heavy metal compounds tend to present significantly higher m values compared with common soil minerals. A powdered mixture behaves ideally (i.e., MA effects are negligible), when the average particle size is less than the critical value dictated by the phase with the highest absorption coefficient (Bish and Post, 1989). The critical particle size was defined by Brindley (1945) as a function of the absorption coefficient, m (1/cm) and the particle diameter, D (cm). Powders were classified into four categories: fine (md, 0.01); medium (0.01, md, 0.1); coarse (0.1, md, 1); and very coarse (md. 1), assuming particle sphericity and uniformity (single size). A correction method for MA effects was developed by Brindley (1945) and can be applied as a post-refinement step in RQA based on the particle size distribution and the average particle size. However, very coarse powders are unsuitable for the Brindley correction. For example, Scarlett et al. (2002) performed quantitative phase analysis on synthetic and natural samples (synthetic bauxite, natural granodiorite). Unground (coarse) material yielded by far the worst results due to MA, and the Brindley correction produced only a slight improvement. Also, Madsen et al. (2001) found that excessive MA correc-

4 490 J. ENVIRON. QUAL., VOL. 36, MARCH APRIL 2007 tion was one of the four major sources of error in RQA, along with the use of unsuitable RIRs, errors in software operation, and user inexperience. Finally, the Brindley approach equates particle size with crystallite size. While this may be true for very fine powders, it is certainly not the case in coarser powders of systems of environmental interest, since weathering reactions cause phase associations. Consequently, the only safe approach to minimize MA effects in any media, including COPR, is to produce adequately fine powders. The critical particle size for COPR is dictated by brownmillerite due to its highest m (Table 1). Classification of the COPR powder as Ffine_ according to Brindley would require pulverization to the submicron level (,0.2 mm), which, however, may introduce potential problems related to overgrinding and/or reactivity. Overgrinding is a problem seldom encountered with hand-pulverization, but its effects have been observed during mechanical pulverization. Prolonged milling can produce very small particles (,,1mm) that are characterized as Fmilling debris_ and add to the amorphicity of the sample (O Connor and Chang, 1986). Jenkins et al. (1986) point out that extremely small crystallite sizes cause peak broadening and reduction in resolution and precision. Softer crystals can be preferentially over-ground and reduced to amorphous debris. Consequently, when there is a large contrast in both m and hardness values, as is the case with COPR (Table 1), a compromise solution has to be found to minimize MA phenomena while retaining mineral crystallinity. Preferred orientation occurs in XRPD powders when completely random orientation of the crystals is not achieved. Randomness enables the various diffraction planes of a crystal to produce proportional reflections, which would otherwise result in misleading diffraction intensities and quantification during RQA. Again, the use of fine-grained powders mitigates the PO problem Table 1. Minerals, chemical formulas, absorption coefficients (m), Moh s hardness, and specific gravity. Mineral Chemical formula m (1/cm) Hardness Gs 2 Amorphous Quintinite Al 2 Mg 4 (OH) 12 (CO 3 ) (H 2 O) 3 Brucite Mg(OH) Ettringite Ca 6 Al 2 (SO 4 ) 3 (OH) 12 I H 2 O Quartz SiO Periclase MgO CAC-14 Ca 4 Al 2 O 6 (CrO 4 )I14H 2 O CAC-12 Ca 4 Al 2 O 6 (CrO 4 )I12H 2 O CAC-9 Ca 4 Al 2 O 6 (CrO 4 )I9H 2 O Sjogrenite (Mg 6 Fe 2 (OH) 16 (CO 3 ) (H 2 O) 4 ) 0.25 Katoite Al 2 O 3 (CaO) 3 (H 2 O) Vaterite CaCO Afwillite Ca 3 (HSiO 4 ) 2 (H 2 O) Calcite CaCO Portlandite Ca(OH) Aragonite CaCO Hydroandradite Ca 3 Fe 2 Si 1.15 O 4.6 (OH) Brownmillerite Ca 2 FeAlO CAC, calcium aluminum chromium oxide hydrate (suffix denotes hydration states). Values obtained from Mineralogical Society of America, Value obtained as density from powder diffraction files (International Centre for Diffraction Data, 2002). (Bish and Post, 1989). There are various models that are incorporated in Rietveld codes to correct preferred orientation during RQA, including the most popular March function, the spherical harmonics model, and the Rietveld-Toraya model (Madsen et al., 2001). Scarlett et al. (2002) reported no significant difference in the accuracy of RQA when either the March function or the spherical harmonics model was applied, while Madsen et al. (2001) reported more accurate results using spherical harmonics. Overall, PO corrections have been found to be much more successful than MA correction with the Brindley method. Choice of Internal Standard Rietveld quantification analysis aims to provide quantitative mineralogical results, based on the relative distribution of the identified crystalline phases in the XRPD pattern. However, a complex sample may usually contain crystalline phases below the XRPD detection limit, disordered crystals that do not provide an adequate signal for detection, and completely amorphous material. For simplicity, the phases that cannot be detected in the XRPD pattern will be defined as amorphous or sample amorphicity. So while the individual amorphous material cannot be distinguished by XRPD and RQA, its total can. To estimate the amorphicity of a specimen with XRPD and obtain true phase quantification, the dried powder is spiked with a known mass of a substance of very high crystallinity, or an internal standard (Jones et al., 2000; De La Torre et al., 2001; Whitfield and Mitchell, 2003). The mixture is then analyzed by XRPD and RQA. The sample amorphicity can be calculated based on the actual mass and the Rietveld calculated weight fraction of the internal standard, enabling true, mass-based comparisons between different samples. The particle size of the internal standard should be comparable to the initial powder to allow adequate homogenization of the two media. Minimization of the MA contrasts (md values) between the initial sample and the internal standard is also preferred. The RIR of the internal standard should be in the range of RIR values of the sample phases to yield comparable intensities at reasonable addition amounts ( 20%). Lastly, the internal standard should not react with the powder. MATERIALS AND METHODS Identical samples of COPR were tested using various preparation procedures to illustrate the impact of the milling type, liquid, and duration on the XRPD and RQA results. Chromite ore processing residue material representing Zone B1 (B1) from SA7 (Chrysochoou and Dermatas, 2006) was used in this study. The mineralogy of B1 COPR is shown in Table 1 along with the corresponding absorption coefficients, mineral hardness (Moh s scale), and specific gravity (Gs) values. Approximately 200 g of B1 COPR were air-dried for 24 h, homogenized, and manually broken down to pass a No. 4 U.S. standard sieve ( 0.5 cm). The material was subsequently divided into 2-g subsamples that were stored separately and used for individual specimen preparation. All analyses were conducted in duplicate, by splitting the pulverized sample and separately preparing each subsample for XRPD analysis. Two

5 DERMATAS ET AL.: X-RAY SAMPLE PREPARATION OF CHROMATE WASTE 491 samples were prepared by hand-pulverization with a ceramic mortar and pestle in dry state, denoted as HP40 and HP400. The HP40 and HP400 samples were processed until each sample mass passed through the No. 40 (0.425 cm) and No. 400 US (38 mm) standard sieves, respectively. A micronizing mill (McCrone Scientific, Ltd) was used for wet mechanical pulverization of the samples using agate grinding elements. Before micromilling, each sample was broken down using a ceramic mortar and pestle to pass through a No. 40 sieve. Cyclohexane (C), isopropanol (I), ethanol (E) (all Fisher Scientific, minimum reagent grade), and deionized water (W) were used as the milling liquids. Table 2 lists the relevant properties of the milling liquids. For each subsample 2 g of solid were milled with 7.5 ml of liquid to comply with the micromill manufacturer guidelines. Milling times of 3, 5, 10, and 15 min were selected to evaluate the influence of grinding on XRPD and RQA (Table 3). After grinding, specimen recovery was accomplished by decanting the slurry in a glass vessel to allow settling of the powdered solids. The clear supernatant liquid was then decanted and the remaining liquid was permitted to evaporate under standard atmospheric conditions until a desiccated powder was obtained. The internal standard was a-corundum (SAWYER), with an absorption coefficient of cm 21 and mean particle size of 1 mm. The amorphous content of a-corundum was estimated at 7% by spiking one sample with a NIST ZnO standard (SRM 674b) on a 50:50 dry weight basis. Corundum was chosen because its peaks did not overlap with major COPR phases, and because its RIR value (1.0) yielded reasonable intensities at 20% addition. Additional details on the choice of internal standard can be found in Pardali (2006). Chromite ore processing residue and corundum powders were blended using a 80:20 ratio by dry weight. The use of 0.8 g COPR with 0.2 g corundum was found to produce the most homogeneous powder. Dry manual mixing was completed with an agate mortar and pestle for 10 min, immediately before the XRPD analysis. The particle sizes of the pulverized COPR specimens and the corundum were examined by the use of a LEO-810 Zeiss scanning electron microscope (SEM). Samples were mounted on aluminum stubs using carbon adhesive pads. Step-scanned XRPD data of the analyzed specimens were collected using a Rigaku DXR 3000 computer-automated diffractometer. The X-ray tube was operated at 40 kv and 40 ma using a diffracted beam graphite-monochromator with Cu radiation (l = Å). The data was collected in the Bragg- Brentano vertical geometry between two-theta values of 5j to 65j with a step size of 0.02j and a count time of 3 s per step. The XRPD patterns were analyzed by the Jade software version 7.1 (MDI, 2005), with reference to the patterns of the International Centre for Diffraction Data (2002) and the Inorganic Crystal Structure Database (2005). Quantitative phase analysis by the Rietveld method was conducted using the whole pattern fitting function of Jade. The refinement strategy was consistent for all analyses presented in this study and elsewhere (Chrysochoou and Dermatas, Table 2. Properties of the milling liquids. Liquid Henry s constant (atm 3 m 3 C Dielectric constant, D Viscosity Phase separation Desiccation time cp h Cyclohexane facile 3 Isopropanol 7.9E difficult 8 Ethanol 5E difficult 24 Water difficult 48 Table 3. Matrix of micromilled samples. Time (min) Liquid Cyclohexane x x x x Isopropanol x x Ethanol x Water x x 2006, Dermatas et al., 2006) to ensure the comparability of RQA results. The strategy was based on the Rietveld refinement guidelines by McCusker et al. (1999), as well as the refinement strategies proposed by Young (1993) and Winburn et al. (2000b), and consultation provided by the Jade developers (Material s Data, Inc.). More specifically: The choice of structural model has been detailed by Chrysochoou and Dermatas (2006). Crystal structure files from the ICSD were used for all phases, except for calcium aluminum chromium oxide hydrate (CACs) (Table 1), as only powder diffraction files (PDF) are available for CACs. A fixed background as a straight baseline was chosen in all cases as peak truncation by polynomial functions was observed. Peaks were modeled using the Pearson-VII function, which gave the lowest residual error (R%) in the peak profile simulation of the NIST Si standard (SRM 640c) analyzed under the same conditions. Lattice constants, scale factors, and full width half maximum (FWHM) were refined for individual phases. Lattice parameters were kept constant when the initial values provided a good fit to reduce the overall number of refinable parameters. The FWHM was restricted to the instrument FWHM (measured at 0.158j using the NIST Si standard) when poor peak resolution or significant peak broadening was observed. The quality of the refinement was judged by the goodnessof-fit S, which is a measure of the ratio of actual residual error vs. ideal residual error. An ideal refinement has therefore an S-value of 1.0. RESULTS AND DISCUSSION After review of the various phenomena in the literature and the experience gained during the experimental timeline, there are three main issues that arise with the production and analysis of powders for XRPD analyses of COPR: Optimizing the ease of powder production (function of pulverization type, milling liquid and time); Minimizing the perturbation of the initial sample and enhancing the reproducibility of the results; and, Comparing the reproducibility of the results with the instrument variability. For all samples, hand pulverization to the No. 40 sieve is taken to be the minimum requirement for XRPD, as one can either continue to hand pulverize the powder, or place it in a micromill for additional pulverization. Accordingly, the production of the HP40 powder was the shortest of all preparations. Conversely, the HP400 powder production required the longest time and highest effort. Hand pulverization of 1.2 g was required to obtain 0.8 g of powder passing the No. 400 sieve, as some of the material was stuck on the mortar and the sieve. The technician effort was significantly higher compared

6 492 J. ENVIRON. QUAL., VOL. 36, MARCH APRIL 2007 Fig. 1. X-ray powder diffraction patterns of duplicate HP40 samples. CAC, calcium aluminum chromium oxide hydrate. to mechanical milling, as brownmillerite, the predominant COPR phase, is a very hard mineral (Table 1) and the initial particle size of COPR is predominantly in the medium sand range. From a milled powder production perspective, cyclohexane proved to be the most advantageous milling liquid because its properties (nonpolarity, high volatility) translated to the most rapid turnaround time (see Table 2). Powder settling occurred almost instantly after decanting the slurry from the milling jar, making phase separation very efficient. The remaining slurry was air-dried for approximately 3 h to produce a desiccated powder (Table 2). Water, ethanol, and isopropanol did not exhibit the same favorable properties due to their higher polarity, viscosity, and lower volatility, the latter of which may have allowed hydration and/or carbonation reactions to occur due to the prolonged exposure to atmospheric conditions during the powder desiccation process (Table 2). Despite the ease of powder production for the HP40 samples, their XRPD patterns (Fig. 1) proved unsuitable for reliable RQA. Strong PO of CACs was observed in one of the duplicate samples, while a strong reflection was observed at a d-spacing of 3.18 Å at the other (possibly PO of albite). While PO may be corrected through the March function in Jade, the lack of structural information and reliable RIR values with PO renders quantification of CACs impossible (see Chrysochoou and Dermatas, 2006, for a detailed discussion on the RIR values of CACs). Scanning electron microscope analyses of COPR samples showed that CACs are large crystals (in the range of mm) and are therefore likely to produce PO in coarse powders (Chrysochoou, 2006). Particle size reduction emerged to be imperative to enable RQA. Preferred orientation was eliminated for CACs in all other samples, while PO observed in portlandite was efficiently corrected using a structural file and the March function. Table 4 summarizes the RQA results for HP400, and micromilled (MM) specimens shown in Table 3. The results are presented as the average of duplicate patterns, plus or minus the range of the results, with the exception of MM5W, for which only one replicate was tested. The goodness-of-fit S was found to be similar for all micromilled samples using liquids with nonpolar moieties. Samples HP400, MM5W, and MM10W had significantly different S values which reflected: high CAC and hydrotalcite peaks caused more complex pattern and thus worse refinement (HP400); extremely high noise and poor peak resolution (MM5W); one large amorphous hump, which was easily modeled, along with fewer pronounced phases (carbonates) yielded an overall easier and more successful refinement (MM10W). The large differences in the S values between the MM5W and the MM10W may be attributed to reactions between COPR and water, not reflected in the S values of other MM samples regardless of the used liquid. The main compounds of interest in COPR are brownmillerite (reactive) and CACs (Cr(VI)-containing). These compounds therefore served, along with the amorphous content, as a basis for comparison of the different sample preparation techniques. Figure 2 compares the brownmillerite and amorphous content for all samples and demonstrated that the results become more consistent in Table 4. Rietveld quantification results of chromium ore processing residue (COPR) samples by sample preparation procedure. Parameter HP400 MM5W MM10W MM5E MM5I MM10I MM3C MM5C MM10C MM15C Goodness-of-fit (S) Amorphous Quintinite Brucite Ettringite Quartz Periclase CAC CAC CAC Sjogrenite Katoite Afwillite Calcite Portlandite Aragonite Hydroandradite Brownmillerite Average and difference of duplicate samples shown in table. HP, hand pulverized; 400, No. of sieve used; MM, micromilled; 3, 5, 10, 15, milling time; C, I, W, E, milling liquid (cyclohexane, isopropanol, water, ethanol); CAC, calcium aluminum chromium oxide hydrate (suffix denotes hydration states, 9, 12, and 14 water molecules).

7 DERMATAS ET AL.: X-RAY SAMPLE PREPARATION OF CHROMATE WASTE 493 Fig. 2. Rietveld quantified brownmillerite and amorphicity results for chromite ore processing residue material. the MM samples with decreasing polarity. Cyclohexane (MMC powders), regardless of the milling time (3 to 15 minutes), produced the most consistent results for the brownmillerite and the amorphicity estimation. Accordingly, to provide focus to the detailed discussion below we will present the results of these powders first. Scanning electron microscope analyses of the MMC powders showed that the minimum milling time (3 min) produced a wider and larger particle size distribution compared with the other MMC powders, as particles.10 mm were observed (Fig. 3a). The SEM analyses of the MM5C to MM15C powders indicated that the increased milling times yielded increasingly uniform particle size distributions and mean particle size powders,10 mm (Fig. 3b through 3d). Among the MMC powders, MA effects were a possible concern for the shortest milling time (3 min), as the MM3C specimen had the widest particle size distribution. The brownmillerite content (the highest adsorbing phase) in the MM3C powder was 2.5% less than the other MMC powders (5 to 15 min) (Table 4). The difference, although not dramatic, indicated that MA effects may have occurred in the MM3C powder due to particle size effects. The relatively constant brownmillerite contents for the remaining MMC powders suggested that MA effects were eliminated for extended milling times. The second most preferable milling liquid from a powder processing perspective was isopropanol (8 h desiccation time, see Table 3). In fact, the MM5I sample closely resembled the MM5C sample. Between themselves, the MMI powders produced comparatively consistent quantification results; however, slightly increased amorphous and reduced brownmillerite contents were observed with increased milling time (5 vs. 10 min), which may indicate slight mineralogical disturbance of the MM10I powder (Fig. 2). Also, the isopropanol specimens may have been fixing CO 2 due to the increased drying period, as reflected by the elevated calcite content in the MM10I powder. Switching from cyclohexane to isopropanol to ethanol (MM5E powder) for the same time interval (5 min) illustrates the potential for increased reactivity in the powder. The brownmillerite content in the MM5E powder was lower (23.2%) compared with the other organic liquids, while the amorphous content was very high (39.9%) (Table 4, Fig. 2). The calcite content was also elevated, likely due to the larger decantation/desiccation times. Prolonged exposure to atmospheric conditions may have allowed hydration and carbonation reactions to occur, which may have affected the mineralogy of the MM5E powder. Whitfield and Mitchell (2003) studied the amorphous content in cements and clinkers and suggested that increased amorphicity may be related to age and storage conditions (duration and conditions of exposure). When water was used as milling liquid, extensive hydration and carbonation reactions occurred in the MM5W and MM10W samples, due in part to the longest desiccation period (48 h, see Table 3). The MM5W and MM10W powders had a brownmillerite content of 11 and 5.4%, respectively, by far the lowest than any other sample (Table 4, Fig. 2), including the HP400 sample. Brownmillerite hydration in coarse-grained COPR is a very slow reaction, which is attributed to the presence of cementitious and iron coatings encapsulating the brownmillerite particles, creating large nodules of 50 to 200 mm size (Chrysochoou, 2006). Particle size reduction presumably destroyed those nodules and increased the reactive surface area, resulting in rapid brownmillerite hydration during milling and dessication. Hydration of brownmillerite most probably resulted in the formation of amorphous iron and aluminum oxides/hydroxides, contributing to the higher amorphicity of the MM5W powder (Table 4, Fig. 2). The high COPR ph also likely lead to rapid CO 2 sequestration during drying, producing the highest calcite content in the MM10W sample (16%) and occurrence of its polymorphs (aragonite and vaterite).

8 494 J. ENVIRON. QUAL., VOL. 36, MARCH APRIL 2007 Fig. 3. Scanning electron microscope analysis of the (a) MM3C, (b) MM5C, (c) MM10C, and (d) MM15C powders (white line equals to 10 mm in all cases). Interestingly, Hillier et al. (2003) used water as milling liquid for wet grinding of COPR with a micronizing mill for 12 min, but reported no evidence of hydration or other reactions during milling. Hillier et al. (2003) indicated that no desiccation period was necessary since spray drying was used, although it was mentioned that ettringite needles were destroyed during specimen preparation. Here, however, the use of water as milling liquid significantly perturbed the mineralogy of the resulting powder to the extent that it could no longer be considered representative of the initial COPR material from the SA7 site. More importantly, the MM5W and MM10W results (vs. the MMC results) do not appear to provide a reliable basis for the assessment of effective remediation alternatives for COPR, mainly due to the significant underestimation of brownmillerite contents. In fact, the HP400 powder, despite systematic MA problems (discussed below) still produced far better results than the MMW powders, suggesting that the role of aqueous reactions is likely the biggest factor affecting the powder preparation process. The HP400 powder was clearly coarser than those mechanically pulverized. The mesh size apparently determined the upper limit of particle size range, but the lower limit was impossible to control by hand pulverization, as illustrated by the particle size distribution of HP400 (Fig. 4). The lack of uniformity in the particle size distribution of the HP400 powder reflects the lack of control on the energy imparted to each particle during grinding and the corresponding particle size reduction. This is reflected in the lower brownmillerite content of the HP400 compared with all MM samples except the MMW samples, presumably caused by MA in the coarser grained sample. The fact that the HP400 sample with no liquid yielded lower brownmillerite content than the MMI and MMC samples further suggests that no to very minimal brownmillerite reactions occurred when using those liquids. The hypothesis that MA caused the differences in the brownmillerite content is also supported by the observation that minerals with lower m values (quintinite, ettringite, quartz, CACs) yielded higher concentration values. This observation would speak for the superiority of mechanical vs. hand pulverization. However, there is one significant reason why hand pulverization may be preferable, and that is the preferential destruction of soft crystals. The comparison of the CAC content of the HP400 sample with all other samples (7.2 Fig. 4. Scanning electron microscope analysis of the HP400 powder (white line equals to 50 mm).

9 DERMATAS ET AL.: X-RAY SAMPLE PREPARATION OF CHROMATE WASTE 495 vs. 1.5%) shows that MA alone cannot account for the difference; it appears that mechanical milling resulted in crystal destruction. Backscattered electron (BSE) images of CAC crystals in various COPR samples showed that CAC crystals are much larger than other phases (10 50 mm) (Chrysochoou, 2006), which may account for their preferential destruction during mechanical milling. Since CACs are important Cr(VI)-binding phases, their quantification is crucial for remediation design purposes. Hand pulverization is therefore necessary as a complementary method to mechanical milling to provide a bound of the concentrations of CACs. The results compiled so far strongly suggest that the MMC approach appears to provide the most objective assessment of COPR mineralogy. To further support this conclusion, two other issues require resolution: Is there an alternative method to provide an objective or absolute quantification of brownmillerite, CACs, and other minerals? What is the relative influence of other sources of error, such as instrument and user variability and choice of refinement strategy? The authors are not aware of any alternative method to establish true quantitative results for COPR mineralogy. Total metal analyses are not particularly helpful, as all major elements are distributed between various minerals, including the phases that do not yield a detectable XRD pattern. Mass balances of the RQA results with acid digestion data have indicated that the amorphous phase primarily consists of calcium and iron, while aluminum and magnesium occur preferentially in crystalline phases; though their inclusion in the amorphous phase cannot be excluded. Furthermore, there is no alternative method known to the authors that could establish the true brownmillerite content. In some cases, the issues can be quite transparent, as in the case of CACs. Quantification of Cr(VI) and mass balance with CACs can provide insight into the maximum CAC content. For example, sample HP400 yielded 7.2% CACs, which corresponds to 0.55% w/w Cr(VI), a concentration that exceeds the total concentration of 0.48% w/w measured by alkaline digestion. While this appears to be an excellent comparison, recent quantification analyses of Cr(VI) in four untreated COPR samples using X-ray absorption near-edge spectroscopy (XANES) showed that alkaline digestion underestimated the total Cr(VI) content by 50% in coarse-grained samples and by 10% on average in fine-grained samples. A definitive conclusion can therefore not be made with regard to the true CAC content. However, the application of both XANES and EXAFS (extended X-ray absorption fourier-transform spectroscopy) appears as a promising alternative to quantify Cr(VI) and CACs, respectively, provided that the necessary standards and access to beam line are available. Separate experiments were completed to test the influence of the instrument variability and repeatability of the sample preparation procedure using a different COPR sample with similar mineralogy. While it is recognized that user variability and the choice of refinement strategy have large impact on the accuracy and precision of RQA, based on both round robin studies (Madsen et al., 2001) and COPR studies (Dermatas and Chrysochoou, 2005), they were not the primary focus of this study. Table 5 shows the results of triplicate scans of the same MM5C powder to gauge instrument variability. The coefficient of variability (COV) values are different for the various minerals and are low (,10%) with the exception of CAC-14, katoite, aragonite, and vaterite. In the case of CAC-14, the variability is due to a change in peak shape, which was reflected in different FWHM and scale factor values. The katoite signal was barely above background and overlapped with hydroandradite and CAC, so that its variability is attributed to slight differences in the quantification algorithm. The COV values are overall higher for minor phases, for which the signalto-noise ratio is low. Finally, the aragonite and vaterite contents increased in each run, which indicates that carbonation took place during the experiment. Overall, and especially with respect to the brownmillerite content, the instrument variability was low and, more importantly, significantly lower than the differences observed between the different sample preparation methods shown in Table 4. Table 6 shows the RQA results for separate MM5C samples prepared according to the same procedures to explore repeatability of the method. The COV values are slightly higher than instrument variability with the exception of CAC-14 (9 vs. 16%) and brownmillerite (2% in both cases). However, standard deviation and COV values are still relatively low. It should also be stressed that the desired precision is a function of the end use of the XRPD results. For example, there is no significant difference between 28 and 30% brownmillerite for all practical purposes when designing a remediation strategy. Rietveld quantification analyses of hundreds of COPR samples have shown that sample variability is significantly higher compared with values presented here, with a standard deviation for brownmillerite on the order of 8% even within the same COPR horizons (Chrysochoou, 2006). Consequently, an absolute error lower than that for a given sample may be considered acceptable. HP400 Table 5. Rietveld quantification results for instrument variability analysis check. Mineral First run Second run Third run Average STDEV COV % Amorphous Brucite Periclase CAC Sjogrenite Katoite Calcite Aragonite Vaterite Portlandite Hydroandradite Brownmillerite All data based on the same MM5C sample, sequential scans; nondetected phases omitted from table. STDEV: standard deviation. COV: coefficient of variability.