On-line X-ray diffraction for quantitative phase analysis: Application in the Portland cement industry

Transcription

1 On-line X-ray diffraction for quantitative phase analysis: Application in the Portland cement industry Nicola V. Y. Scarlett and Ian C. Madsen CSIRO Minerals, Box 312, Clayton South, Victoria 3169, Australia Con Manias and David Retallack Fuel and Combustion Technology Pty. Ltd. (FCT), 20 Stirling Street, Thebarton SA 5031, Australia (Received 8 September 2000; accepted 12 October 2000) The aim of this work was to design, construct, install, and commission an on-line, X-ray diffraction (XRD) analyzer capable of continuously monitoring phase abundances for use in process plant control. This has been achieved through a joint project between CSIRO Minerals and Fuel & Combustion Technology Pty. Ltd. with an instrument designed for use in a Portland cement manufacturing plant. Key factors in tailoring such an instrument to the cement industry were (i) the handling and presentation of a dry sample and (ii) the development of an analytical method suitable for the complex suite of phases contained within Portland cement. The instrument incorporates continuous flow of sample through the diffractometer using a purpose-built sample presentation stage. The XRD data are collected simultaneously using a wide range (120 20) position sensitive detector, thus enabling rapid collection of the full diffraction pattern. The data are then analyzed using a Rietveld analysis method to obtain a quantitative estimate of each of the phases present. The instrument is controlled by a PC linked to the diffractometer through a purpose built interface. The phase abundance information is then transmitted to the central computer in the cement plant where it can be used for the control of mill parameters such as temperature and retention times as well as gypsum feed rate International Centre for Diffraction Data. [DOI: ] I. INTRODUCTION Many processing industries require routine analysis at various stages of processing as a means of controlling the quality of their products. Frequently, this consists of an ' 'offline" chemical analysis obtained by sub-sampling the processing line and sending the material to the site laboratory for analysis. This "batch" mode of operation often incurs sampling problems and poor turnaround times, both of which can limit the usefulness of the results for plant control. In many cases, knowledge of the phase, rather than the elemental, abundances is required to assess the processing conditions. In the Portland cement industry, phase abundance determination has traditionally been achieved in an "offline" manner either by: Optical microscopy using point counting techniques (Campbell, 1986); or Use of the Bogue (Bogue, 1929) method where phase abundances are derived by reduction of the chemical analyses assuming a detailed composition for each phase. Optical microscopy is very labor intensive and can be severely affected by sampling errors. The Bogue method often produces considerable error due to variation in the phase composition resulting from changes in the feed-stock andor processing conditions. Further discussion of the errors in these methods can be found in Madsen and Scarlett (1999) and references therein. Both optical microscopy and the Bogue method are considered to be too slow to be directly incorporated into a plant control circuit since the results may not be available for several hours. An alternative is to use X-ray diffraction methods with a continuously moving stream of material and to apply the Rietveld method to extract the quantitative phase analysis. This "on-line" determination of phase abundance has some significant advantages, namely: XRD determines the phase abundances directly from the crystallography of the phases present, not by inference from other measurements (e.g., chemical analysis). The moving sample significantly reduces the problems associated with particle statistics encountered in conventional "stationary sample" XRD instruments. Therefore, larger particle sizes can be tolerated without affecting the reliability of peak intensity measurements, and hence the phase abundances. The turn-around time for the analysis is very rapid (of the order of 1-10 min), thus enabling their direct use in plant control. CSIRO Minerals and Fuel & Combustion Technology Pty. Ltd. have collaborated in the design and construction of an on-line XRD instrument for the cement industry. The following steps were involved in the development of the instrument: (1) Selection of XRD instrument geometry and components required; (2) Determination of the applicability of the phase suite to quantitative phase analysis via the Rietveld method and development of a suitably accurate and fast refinement strategy; (3) Devising a method for rapid data collection; (4) Design and construction of a sample presenter for the continuous feeding of a sample suitable for X-ray diffraction studies; and 71 Powder Diffraction 16 (2), June (2)71 10$ JCPDS-ICDD 71

2 TABLE I. Phases present in Portland cement clinker. Phase Nominal formula Typical concentration (wt. %) Comments C4AF C:a 4 (Al ;t Fe, -,) 4 O 10 Lime Ca 3 SiO 5 Ca 2 SiO 4 Ca 3 Al 2 O 6 Ca, _,Mg.O Seven known polymorphs three present in clinker Five known polymorphs J- known to be present in clinker, but some evidence for presence of another two Only one polymorph, but composition varies (0.0<;t<0.5) Two polymorphs cubic and orthorhombic Mg substitution for Ca (0.0<;t<0.02) Hydrates rapidly to portlandite Ca(OH) 2 (5) Design and coding of software for (i) remote control of the instrument, (ii) data acquisition and analysis, and (iii) reporting of results to the site computer. II. EXPERIMENTAL A. Cement analysis 1. Portland cement production and phases produced The methodology of production for Portland cement is well documented (Taylor, 1990) and will not be repeated in detail here. In summary the process involves: Feed material consisting mainly of CaO, SiO 2, A1 2 O 3, and Fe 2 O 3 is prepared by milling predetermined quantities of limestone, shale, sand and iron oxide. The mixture is heated in a rotating kiln at approximately 1400 C. The limestone decomposes and calcium silicate, calcium-aluminate, and calcium-alumino-ferrite phases are formed in "clinkers" of approximately 1-3 cm in diameter. The composition of the feed materials as well as kiln conditions can affect the final clinker properties and composition. Table I describes the clinker phases and their nominal compositions. Portland cement is prepared by grinding the clinkers in a ball mill with approximately 5 wt. % each of gypsum (CaSO 4.2H 2 O) and limestone. The mill temperature rises to about 130 C during grinding, causing partial decomposition of the gypsum to form hemihydrate (CaSO 4.0.5H 2 O) and occasionally anhydrite (CaSO 4 ). Table II gives the additional phases present in Portland cement. At two crucial stages in the manufacture, phase abundance estimations can be used to provide feedback for plant control. First, analysis of the clinker material provides information for control of kiln parameters such as temperature profile, residence times, and excess oxygen. Second, analysis of Portland cement provides quality assessment of the finished product as it is produced rather than relying on testing of material at the time of dispatch to the customer. The measurement of the mineralogy allows (i) control of the setting time (estimated from the gypsum, hemihydrate and ratios), and (ii) control of the weighfeeders for gypsum, limestone, and clinker in the feed to the mill. 2. Phase abundance determination procedure Traditional quantitative phase analysis using XRD has been carried out using single peak methods based upon either peak heights or integrated intensities (Klug and Alexander, 1974; Cullity, 1978; Jenkins and Snyder, 1996). The application of such methods to the quantitative analysis of cement and clinker phases has not been entirely successful in the past (Aldridge, 1982). This is due to the high degree of overlap between the main peaks of the major phases leaving few free-standing peaks available for the measurement of integrated intensities. In order to avoid the overlap problem, relatively weak peaks are often selected for analysis, leading to large counting errors in the determination of peak intensity. Figure 1 shows an observed XRD pattern for clinker plus the individual components (calculated to represent typical clinker composition) for each of the major phases and clearly demonstrates the high degree of overlap in the region 31 to (CuKa) where most phases have their strongest reflections. Additional complications in the XRD analysis of Portland cement include: Most of the phases display variable amounts of atomic substitution (solid solution) which can have a substantial influence on the measured peak positions for a particular sample. Variation of the degree of crystallinity of the phases can affect the width and shape and hence the degree of overlap between the observed diffraction peaks. TABLE II. Phases present in Portland cement in addition to those present in clinker. Phase Nominal formula Typical concentration (wt. %) Comments Gypsum Hemihydrate Anhydrite Calcite Quartz CaSO 4.2H 2 O CaSO 4.;tH 2 O CaSO 4 CaCO 3 SiO Exhibits severe preferred orientation Hydration state varies (0.3<J:<0.8) Exhibits preferred orientation 72 Powder Diffr., Vol. 16, No. 2, June 2001 Scarlett et al. 72

3 f Observed Lime 1 \J! S A K \ - \ V ( ) A C4AF I 1 V; \ Figure 1. Observed XRD pattern (CuKa) for a sample of Portland cement clinker showing the individual components (calculated to represent their abundance in a typical clinker) for each of the major phases. Note that the y-axis is depicted as the square root of the actual intensities to improve the clarity for the minor phases. (as defined in Table I) exhibits strong preferred orientation about the (001) crystallographic direction. This affects not only its own relative peak intensities, but also the derived peak intensities of overlapping phases, especially. In recent years it has been recognized that peak overlap problems (and, to a lesser extent, preferred orientation) can be largely solved by using the entire diffraction pattern for analysis, rather than a small number of preselected, freestanding peaks. The Rietveld method (Rietveld, 1969; Young, 1993) uses a model which includes the crystal structure for each phase, the pattern background, and peak width and shape parameters to generate a calculated diffraction pattern which is then compared with observed diffraction data. Parameters in the model are varied through a least-squares process to minimize the difference between the calculated and observed patterns. For each phase a scale factor is included to adjust the intensity or "height" of the calculated pattern. In a multiphase sample consisting of n phases, there is a relatively simple relationship between the Rietveld scale factor and the relative abundance of phase i [Eq. (1)]. This technique was first proposed by Hill (1983) and further developed and tested by Hill and Howard (1987) and is given by: Si(ZMV) t (1) where w, = weight fraction of phase, 5 = refined Rietveld scale factor, ZMV'= the "calibration constant" where Z is the number of formula units per unit cell, M is the mass of the formula unit, and V is the volume of the unit cell. The Rietveld method is now widely used (Madsen et al., 2001) for phase abundance determination from diffraction data since no calibration is required; in this case knowledge of the crystal structure provides the calibration constant. It should be noted that the so-called ZMV method provides a dynamic calibration throughout the refinement process but provides only relative phase abundance, i.e., the analyzed phases are normalized to 100%. The advantages of the Rietveld method for the quantitative phase analysis of Portland cement are: The use of the entire diffraction pattern (where tens or hundreds of peaks may contribute to the analysis) rather than just one or two peaks as in conventional methods. The potential for refinement of the crystal chemistry of the component phases the structural parameters may then contain information which can be correlated to processing conditions. The reduction of peak overlap difficulties since the observed data are compared with a calculated pattern which is the sum of the component phase contributions. However, some of the disadvantages of the method are the requirements for: Detailed crystallographic knowledge of all phases present. Not all of the structures are well defined for the clinker phases, i.e., reported structures may not calculate intensities which match those observed. Data which cover a wide angular range need to be collected. Potentially long computational times. Many of the phases present in cement and clinker have complex crystal structures that generate many peaks in the calculated pattern. The computational time is proportional to the number of peaks in the calculation. In addition, some phases are polymorphic and may be present as a mixture of two or more forms within a single material. Calculation of relative rather than absolute phase abundances. This does not account for the presence of any amorphous component in the sample stream. In the case of the Portland cement industry the amount of amorphous content was deemed to be of an insignificant level and thus quantification of relative phase abundances is adequate for plant control. If large amounts of amorphous material were a regular component of another sample stream it may be possible to include an assessment of the amount by measurement of the area of the amorphous peak. In many industrial applications the appearance of trends in the relative proportions of phases is still useful for plant control even if absolute quantification is not possible. The need for long data sets and computational times has the potential to reduce the efficacy of the method as a tool for rapid feedback of results for the purpose of plant control. Modified Rietveld approach. In order to address the difficulties of ill-defined structures and long computational times, a modified Rietveld approach (Taylor and Rui, 1992) has been taken to the analysis of Portland cement. Where the conventional Rietveld method utilizes crystal structure information to generate peak intensities, the modified approach uses intensities based, in part, on intensities observed in "real" materials. This technique is particularly useful for the analysis of since the reported structural information does generate peak intensities which accurately reflect those observed in real clinker material. In summary, the method- 73 Powder Diffr., Vol. 16, No. 2, June 2001 On-line X-ray diffraction for quantitative phase 73

4 ology (described more fully in Taylor and Rui, 1992; Madsen and Scarlett, 1999) comprises: Calculate the structure factors using the partially known structure and write to a text file. Manually modify the structure factors according to the observed intensities to obtain a better fit during Rietveld refinement. Adjust the absolute scale of the modified structure factors to ensure compliance of the ZMV relationship used for derivation of phase abundances. During analysis, read the structure factors from the file rather than generate them from the crystal structure. To speed up processing it is also possible to read in unmodified values of the structure factors rather than repeatedly calculate them from the structure throughout the refinement. In the case of the Portland cement detailed here, all phases were read in from structure factor files from which the very weak reflections (less than 0.5% of the maximum for the phase) had been omitted to further speed the calculations. Although detailed structural knowledge is not required for this approach, it is necessary to know the unit cell of the phase in question (i.e., all peaks must be able to be indexed) to adjust the peak positions if necessary. In this work only the structure factors for were modified (based on the average intensities observed from several "real" clinker samples) in order to provide the best match between the calculated and observed XRD patterns. Madsen and Scarlett (1999) have noted that the XRD pattern of cement clinker exhibits a peak at approximately (CuKa) which can be attributed to (ICDD-PDF, 1995 card number ). In addition, the peak can be indexed using the structure of Mumme (1995) but is not generated with significant intensity by any of the structures used for pattern calculation. Modified structure factors for this peak have also been included in the modified file. While this approach was effective in improving the general level of fit between the observed and calculated patterns in these samples, additional modifications to the structure factor files may be required for clinker or cement samples from other sources. Changes to the source of raw materials or to kiln operating conditions at a particular location may also produce minor changes in the observed peak intensities. B. Instrument design 1. X-ray diffractometer The diffractometer used for the XRD analyzer was composed of (i) an Inel CPS 120 curved detector plus electronic control panel, (ii) interface card, (iii) X-ray generator with RS232 interface, (iv) a graphite, incident beam monochromator (v) Co target tube, and (vi) a purpose-built instrument cabinet. It is the use of the curved detector, which allows the simultaneous collection of of data, that forms the basis of the on-line XRD analyzer. The resolution of this detector is of the order of It is the only one currently available capable of simultaneously collecting the angular range required for this application. Figure 2 shows the diffractometer with the sample presenter in place. There remain some problems inherent in the use of the curved detector in this application. The most important is Figure 2. The upper cabinet of the on-line XRD instrument showing the continuous feed sample presenter. that the diffraction patterns are reported in terms of channel numbers rather than diffraction angle required by the Rietveld program. This has been overcome by devising a calibration routine that is relatively easy to use in a plant environment. However, small residual errors remain in the reported 20 values and must be accommodated by the Rietveld refinement software (Coelho et al, 1997). Additional problems are caused by the inability to place a postdiffraction monochromator in the instrument, resulting in an increase in pattern background due to sample fluorescence (Figure 3). This difficulty can be partially solved by using CoKa radiation to eliminate the generation of FeA"a fluores- a Intensity (coun _ 4000 $ A i I I! i, i H 26 f) Figure 3. X-ray diffraction data collected using the on-line analyzer using total collection times of 1 min (A) and 10 min (B) duration. Note (i) the relatively high pattern background due to sample fluorescence and (ii) the general improvement in the definition of the minor phases above the pattern background with increased counting time. 74 Powder Diffr., Vol. 16, No. 2, June 2001 Scarlett et al. 74

5 \ :. 9 \, Figure 4. Sample presenter for on-line XRD cement analyzer showing the sample feed tube (A), the body of the presenter (B), and the sample surface point, thus providing a representative sample for analysis. This should be compared to conventional laboratory based measurements (using X-ray fluorescence spectrometry) where only about 5 kg of material is sampled over a 2-4 h period and then sub-sampled down to a few grams for analysis. The sample presenter provides a smooth, flat sample surface of constant height by means of a roller and knife-edge system. Following irradiation, the sample is withdrawn by means of a scraper and vacuum system that returns the material to the cement storage bin. The rate at which the sample passes through the diffractometer can be adjusted to vary the total amount of sample examined in each analysis. Figure 4 shows the sample presenter mounted on the diffractometer. The ability to examine a continuous stream of sample in this manner significantly increases the number of crystallites examined during the data collection, and hence improves the precision of the quantitative phase analyses. (C). 3. Softwareinstrument controller cence. However, the presence of high concentrations of calcium in the samples still causes problems due to CaKa fluorescence. While much of the CaKa is absorbed in the air gap between the sample and the detector, there is still sufficient to increase the pattern background and hence increase the lower limits of detection for the minor phases. 2. Sample feeder and presenter The sample handling devices have been purpose-built to provide a continuous flow of material from the cement mill, through the diffractometer, and back to the cement hopper. While the total amount of cement manufactured is some 75 tonnes per hour, only a fraction of this (several kgh) is passed through the XRD analyzer. The grinding process in the mill ensures uniform mixing of material at the sampling The operation of the on-line diffractometer is controlled remotely by a PC. Purpose-written software (Figure 5) controls all aspects of the operation of the instrument including power onoff, checking of instrument status parameters, calibration of the detector, control of the XRD instrument parameters, collection and analysis of data, and the reporting of results to the plant operating room. Data quality can be controlled through the use of a data collection regime in which single data sets (typically of 60 s duration) are collected and several (typically ten) are summed prior to Rietveld analysis. The benefits of this regime include (i) improvement of the X-ray counting statistics through the addition of multiple data sets, and (ii) the generation of rapid analyses (every 1 min) which are based on the running sum of the previous ten data sets. The software Eh»» U * Cat p imtiacwftr [ i PHASE WSUlT MM c Brila ^ Figure 5. The main screen for the software interface for the on-line XRD cement analyzer. Features include (i) display of the number of data sets required for analysis and the number already collected, (ii) the elapsed time for the current data set, (iii) controls to power the XRD analyzer up and down as well as to start data collection and Rietveld analysis, (iv) display of the quantitative phase abundances and derived oxide contents for the most recently analyzed data set, and (v) the status of various sections of the instrument. q-77 M»* F«rt»b ots AkJimtro '43f. Almnta-c J0O1 Una UXO FMInto '343 Sjwum «--.(6!4 Mvdto Cafcto KP4 OiMb L *4 4 H.'t RFKKX?*= U) f.4,--.;. ua I8f37 ALMMl MESSAGE MLgr.«l»nM«ml y Mctwolr Wad Dro«nii 75 Powder Diffr., Vol. 16, No. 2, June On-line X-ray diffraction for quantitative phase 75

6 Figure 6. The on-line XRD cement analyzer installed in the purpose-built instrument room adjacent to the cement mill. The upper cabinet houses the XRD instrumentation and sample presenter, while the lower cabinet houses the generator, detector electronics, and computer. allows the setting of both the data collection time and the number of data sets prior to analysis. The instrument described here employs a flat graphite, incident beam monochromator. Data quality could be improved by the inclusion of a multi-layer mirror in the incident beam path to provide an increase in pattern intensity of about 5 to 10 times. Adequate data could then be collected in a total of 1-2 min. 4. Instrument construction and installation The XRD instrument is housed in a purpose-built cabinet (Figure 6) comprised of two sections that can be sealed to exclude dust and separated for transportation. The upper section houses the INEL XRD instrument and sample presenter, while the lower section houses the detector electronics, X-ray generator, instrument interface, and computer. The lower cabinet is equipped with a heat exchanger to remove the excess heat generated by the electronics. The upper cabinet section has been designed to minimize the risks of radiation leakage, with special attention paid to the conditions likely to be encountered in a plant installation. III. RESULTS AND DISCUSSION A. Verification of analytical method One of the difficulties with the analysis of cement and clinker is in establishing a "true" quantitative phase analysis 76 Powder Diffr., Vol. 16, No. 2, June 2001 (QPA) with which to compare results from XRD phase analysis. The industry standard methods of point counting or the Bogue method both suffer severe limitations (Madsen and Scarlett, 1999). However, given that the Bogue method is the most commonly used benchmark, it was necessary from an industrial point of view to relate the method described here to it. More accurate verification of the analytical method was obtained by comparing "reduced oxide" values calculated from the Rietveld QPA and the detailed phase compositions obtained from microprobe analysis against the oxide values measured by X-ray fluorescence. A series of samples of plant material were provided along with their relevant Bogue and XRF analyses for verification work. XRD data were collected on these samples using a traditional Bragg-Brentano geometry, scanning diffractometer incorporating a post-diffraction monochromator to eliminate sample fluorescence. This provided high quality data sets with good counting statistics and high peak-tobackground ratios. The results of QPA using the modified Rietveld approach are compared with the Bogue results in Figure 7, while a comparison of the oxide contents from XRF and the XRD "reduced oxides" is shown in Figure 8. Table III shows the standard deviation (s.d.) for each method derived from the covariance between the (i) XRD and Bogue, and (ii) XRD and XRF oxide estimates using the method of Grubbs (1948). Also included is the mean difference between the estimates for each of the major phases (Bogue XRD) and oxides (XRF XRD reduced oxides). The individual s.d.'s provide a measure of the precision of each analysis method, while the mean difference provides an indication of the overall accuracy (incorporating the bias present in each method). For the most abundant phases ( and ), the high s.d. values reflect relatively poor precision in the determination, with the Bogue being much poorer than XRD. For the other phases (C4AF and ) both methods contribute approximately equally to the overall error estimate. For all phases, the large values of the average difference indicates relatively poor accuracy when attempting to use Bogue results as a point of comparison for the XRD analyses. For all of the major oxide components, the s.d. values are generally much lower than for the phase data with the XRF estimates and those derived from XRD contributing approximately equally to the overall error estimate. In addition, the values for average difference for the oxides are much lower than those given for the phase abundances. These results show that the comparison of reduced oxides calculated from XRD results with XRF measurements is a more reliable measure of success of the XRD analysis than comparison of XRD phase abundances with the Bogue method. 1. Sensitivity of lime measurement Lime (CaO) is present in clinker at a level of approximately wt. % and is a useful indicator for kiln operating conditions. Ideally, an accuracy of ±0.1 wt. % is required in the analysis, but ±0.2 wt. % is useful for obtaining trends. Verification of the accuracy of lime measurement by XRD was achieved by preparing a synthetic sample consisting of clinker with various known lime additions. XRD data were collected using a conventional, laboratory based scanscarlett et al. 76

7 c n a 40 a. S CD » C4AF x -0 1: XRD Phase Abundance (wt%) 5 10 XRD Phase Abundance (wt%) 15 Figure 7. Comparison of quantitative phase abundance determination for plant samples using the modified Rietveld approach with that from Bouge analysis. "A" shows all phases present simultaneously. "B" is an expansion of "A" which shows the minor phases in greater detail. ning diffractometer incorporating a dry nitrogen atmosphere and a sample stage heated to 110 C to prevent the hydration of lime to portlandite. The results of the analyses are compared with the known additions in Table IV. The results indicate that lime can be measured to an accuracy of about ±0.2 wt. % at levels up to 2 wt. % which should be sufficient for obtaining trends in the lime content. Future developments will aim to improve the level of accuracy of the method. The estimation of the lime (CaO) content will only be effective for freshly manufactured clinker since lime readily hydrates to portlandite [Ca(OH) 2 ]. It is unlikely that lime will be detected in finished cement since extended storage periods for the clinker and the use of a water spray to control mill temperature will ensure that all of the lime has hydrated. The presence of portlandite provides an indication of pre-hydration (related to problems with cement mill water sprays or clinker storage) and is a factor used in the prediction of cement strength (Manias et al, 2000). While portlandite can be included in the quantitative phase analysis, it should be noted that the crystallite size of portlandite is relatively small. This results in broad diffraction peaks that are not readily observed above the pattern background unless the counting statistics are adequate. The instrument operators A o B I. a u IS o c I S Lice x XRD "Reduced Oxide" Abundance (wt%) CaO SiO2 * AI2O3 x Fe2O3 MgO + Na2O - K2O 1:1 XRIF Oxide Abundar 1 n XRD "Reduced Oxide" Abundance (wt%) CaO SiO2 A AI2O3 x Fe2O3 MgO + Na2O - K2O 1:1 Figure 8. Comparison of reduced oxide values derived from XRD phase abundances with analyzed values from X-ray fluorescence for the same suite of samples as shown in Figure 7. Once again "A" shows all phases present simultaneously and "B" is an expansion of "A" which shows the minor phases in greater detail. 77 Powder Diffr., Vol. 16, No. 2, June 2001 On-line X-ray diffraction for quantitative phase 77

8 TABLE III. Estimates of the covariance and average difference for (i) Bogue and XRD for the major phases and (ii) XRF oxide estimates and the "reduced oxides" derived from XRD. The average difference is defined as the mean of (Bogue XRD phase estimate) or (XRF XRD reduced oxide) for each determination. A 16 i Standard deviation Derived from covariance C4AF Phase C4AF Bogue XRD Average difference o 8o O 9 o 6 ID O 4 Standard deviation Derived from covariance 17:00 19:00 21:00 23:00 1:00 3:00 5:00 7:00 Oxide XRF oxides XRD reduced oxides Average difference Time (hours:minutes) CaO SiO 2 A1 2 O 3 Fe 2 O 3 MgO must consider the importance of counting statistics when selecting the data collection times. B. Laboratory based stability testing Prior to the installation of the on-line XRD analyzer, extensive laboratory based testing was conducted to determine the stability of the instrument. Some of these tests involved the collection of data from a recirculating sample over a period of 15 h. Data sets of 60 s duration were collected continuously and analyses carried out on the summation of ten sets. The result of this was the production of a QPA result every 1 min which is the average of the previous ten minutes of data collection. Figure 9 shows the results of these stability trials for the major phases contained within the Portland cement. Table V shows the means, standard deviations, and % relative standard deviations (%RSD) for these results. The maximum deviation over the course of 15 h was 3.70% relative in the gypsum (for example) which was present at a level of about 3 wt. %. This indicates good reproducibility and internal stability of the instrument in a laboratory environment. C. Plant based testing Testing similar to that carried out in the laboratory was also carried out in the plant as part of the on-line XRD in- TABLE IV. Results of quantitative phase analysis for known lime additions to a sample of cement clinker. The figures in brackets for the analyzed values represent the errors derived from the errors in the Rietveld scale factors. The line of best fit is given by ana!yzed=0.8994*weighed , K 2 = Weighed (wt. %) Analyzed (wt. %) Bias (wt. %) s E I $ 17:00 19:00 21:00 23:00 1:00 3:00 Time (hours:minutes) Figure 9. Results of laboratory based stability testing for on-line cement analyzer over a 15 h period for (A), C4AF,, and and (B) Gypsum and Hemihydrate. For the duration of the test, the same cement sample was allowed to remain on the sample presenter. For (A), the left hand scale refers to the amount of present, while the right hand scale refers to the amount of all other phases. For (B), the left hand scale refers to the amount of gypsum present, while the right hand scale refers to the amount hemihydrate. strument's commissioning process. In this instance, the cement manufacturers were interested in trialing faster turnaround times than those considered in the laboratory so analyses were carried out on single data sets of 80 s duration. The sample was fed continuously from the cement mill and data were collected over a period of 8.5 h. The QPA results for the major phases obtained during plant testing are shown in Figure 10 and Table VI. Note that the cement used in this trial was not the same as that used in the laboratory trials, so the magnitude of the results will be slightly different. TABLE V. Means, standard deviations and % relative standard deviations (RSD) for laboratory based stability trials. Individual analyses were based on the rolling sum of 10X60 second data sets (0.10) 0.31(0.10) 1.13(0.12) 1.78(0.14) 4.61(0.16) «= 815 Mean S.D. %RSD C4AF Gypsum Hemihydrate Powder Diffr., Vol. 16, No. 2, June 2001 Scarlett et al. 78

9 A. the material fed continuously from the cement mill. While correlation of the QPA results with actual mill operating conditions has not been performed to date, the variations observed in the gypsumhemihydrate values (Figure 10) could provide valuable data for control of the mill operating conditions. S 50 V) o 45 0 ' C4AF 8:30 9:30 10:30 11:30 12:30 13:30 14:30 15:30 16:30 17:30 B r * * -:.?r*x*,' * Time (hours:minutes) % - a..... '-;. :\:- Gypsum. : - '." '. '. "... Hemihydrate 8:30 9:30 10:30 11: :30 14:30 15:30 16:30 17:30 Time (houre:minutes) Figure 10. Results of plant based stability testing for the on-line cement analyzer over an 8 h period for,, C4AF, and (A) and Gypsums and Hemihydrate (B). During testing, the sample was fed continuously from the cement mill. For (A), the left hand scale refers to the amount of present, while the right hand scale refers to the amount of all other phases. For (B), the left hand scale refers to the amount of gypsum present, while the right hand scale refers to the amount hemihydrate. The solid lines represent the mean of each set of analyses in this diagram while the dotted lines represent ±2 esds derived from the laboratory based testing and show the limits of variability that can be expected from instrumental variation alone. In addition to quantitative phase analyses, Figure 10 shows the mean for the plant based results as well as lines representing ±2 estimated standard deviations (esd) derived from the laboratory based testing. The ±2 esd lines indicate the limit of variation that could be attributed to instrument stability alone. For the sulphate phases (gypsum and hemihydrate) variations are apparent which exceed these limits and could be attributed to variability in the composition of TABLE VI. Means, standard deviations, and % relative standard deviations (%RDS) for plant based trials. Individual analyses were based on single 80 second data sets. n = 315 Mean S.D. %RSD C4AF Gypsum o I s g o 5" Hemihydrate IV. CONCLUSIONS An on-line, X-ray diffraction instrument capable of continuously monitoring phase abundances has been constructed and installed in an operational cement plant. This has been achieved through (i) the use of a curved, position sensitive detector to simultaneously collect wide-range diffraction data, (ii) the incorporation of a continuous feed mechanism for sample presentation, and (iii) the use of a modified Rietveld approach for quantitative phase analysis, thus providing more accurate analyses than conventional methods. The continuous nature of the sample feed significantly reduces the problems associated with obtaining a representative sample by more traditional methods of sampling. An additional benefit of the use of a moving sample is the increase in the total number of particles examined during the data collection. This leads to increased accuracy in the peak intensities and hence the quantitative phase analyses. In its current configuration, this instrument can be used to analyze any multi-phase system that is amenable to analysis by powder X-ray diffraction. The instrument can be rapidly reconfigured to allow a "single sample" mode of operation for the periodic investigation of smaller samples for research purposes. Currently, the sample feed system is limited to dry feed materials but work is in progress to allow for the continuous analysis of slurry samples resulting from mineral processing operations. ACKNOWLEDGMENTS Edmund Schneider is gratefully acknowledged for his input to the design and construction of the sample feed and presentation system. The staff of the Birkenhead works of Adelaide Brighton Management Ltd. are gratefully acknowledged for their assistance during installation and commissioning of the XRD cement analyzer. Dr. Cheryl Lim and Dr. Nick Cutmore of CSIRO Minerals are gratefully acknowledged for their reviewing of an early version of the manuscript. Aldridge, L. P. (1982). "Accuracy and precision of phase analysis in Portland cement by Bogue, microscopic and X-ray diffraction methods," Cem. Concr. Res. 12, Bogue, R. H. (1929). "Calculation of the compounds in Portland cement," Ind. Eng. Chem. 1, Campbell, D. H. (1986). "Microscopical examination and interpretation of Portland cement and clinker," Portland Cement Association, USA, Library of Congress catalogue card number Coelho, A. A., Madsen, I. C, and Cheary, R. W. (1997). "A New Rietveld refinement program using a fundamental parameters approach to synthesizing line profiles," Proceedings of Crystal XX, The Twentieth Meeting of the Society of Crystallographers in Australia, April 2-5, 1997, Queenstown, New Zealand. Cullity, B. D. (1978). Elements of X-ray Diffraction, 2nd ed. (Addison Wesley, New York). Grubbs, F. E. (1948). "On estimating precision of measuring instruments and product variability," J. Am. Stat. Assoc. 43, Powder Diffr., Vol. 16, No. 2, June 2001 On-line X-ray diffraction for quantitative phase 79

10 Hill, R. J. (1983). "Calculated X-ray powder diffraction data for phases encountered in leadacid battery plates," J. Power Sources 9, Hill, R. J., and Howard, C. J. (1987). "Quantitative phase analysis from neutron powder diffraction data using the Rietveld method," J. Appl. Crystallogr. 20, ICDD-PDF (1995). "The database of diffraction data compiled by the JCPDS International Centre for Diffraction Data." Jenkins, R., and Snyder, R. (1996). Introduction to Powder Diffractometry (Wiley, New York). Klug, H. P., and Alexander, L. E. (1974). X-ray Diffraction Procedures for Polycrystalline Materials, 2nd ed. (Wiley, New York). Madsen, I. C, and Scarlett, N. V. Y. (1999). "Cement: Quantitative phase analysis of Portland cement clinker," in Industrial Applications of X-ray Diffraction, edited by F. H. Chung and D. K. Smith (Marcel Dekker, New York). Madsen, I. C, Scarlett, N. V. Y., Cranswick, L. M. D., and Lwin, T. (2001). "Outcomes of the International Union of Crystallography Commission on Powder Diffraction Round Robin on Quantitative Phase Analysis: Samples 1A to 1H," J. Appl. Crystallogr. (submitted). Manias, C, Retallack, D., and Madsen, I. (2000). "XRD for on-line analysis and control," World Cem Mumme, W. G. (1995). "Crystal structure of tricalcium silicate from a Portland cement clinker and its application to quantitative XRD analysis," Neues Jahrb. Mineral., Abh. 169, Rietveld, H. M. (1969). "A profile refinement method for nuclear and magnetic structures," J. Appl. Crystallogr. 2, Taylor, H. F. W. (1990). Cement Chemistry (Academic, London). Taylor, J. C, and Rui, Z. (1992). "Simultaneous use of observed and calculated standard profiles in quantitative XRD analysis of minerals by the multiphase Rietveld method: The determination of pseudorutile in mineral sands products," Powder Diffr. 7, Young, R. A. (1993). The Rietveld Method (Oxford University Press, Oxford). 80 Powder Diffr., Vol. 16, No. 2, June 2001 Scarlett et at. 80