THE POSSIBILITY OF USING HANDHELD XRF IN CEMENT APPLICATIONS Dr. Michelle Cameron, Bruker Elemental, Kennewick, WA Denver X-ray Conference 2011 ABSTRACT Handheld XRF may become a useful and cost-saving tool in the cement industry, especially for measuring raw materials and making quick decisions in the field. Calibration methods are explained and results are presented for a Bruker S1 TURBO SD LE used to measure limestone and related materials used in the cement industry. Results are discussed for measurements of powders, pressed pellets, and unprepared rock samples. Variations in calibration assumptions are also discussed, which allow better tailoring of a calibration to measurement of different matrices. INTRODUCTION Laboratory and benchtop x-ray diffraction (XRD) and x-ray fluorescence (XRF) are used regularly as part of the quality control process in cement plants. These have all but replaced traditional wet chemistry methods in this process. X-ray-based methods are considered as rapid analysis tools to ensure quality final product, and are qualified by demonstrating compliance to the requirements in ASTM C-114. They are also used in other parts of the cement-making process to monitor raw materials and clinker. Use of laboratory XRD and XRF requires significant sample preparation. This is important for XRF because low energy x-rays from elements like Mg, Al, and Si are very weak and don't travel far in air or in a solid matrix. Table 1 shows the path lengths of selected elements through air and through a silica matrix. Even Ca, which is a "heavier" element in the cement process, only has a path length of 27μm. Since only the first 30μm of the sample is being measured, ensuring homogeneity of the sample is essential for accurate measurement. Element Air (cm) Silica (microns) Be 0.049 0.1 C 0.428 0.5 O 0.147 2.1 Na 0.595 3.1 Mg 0.986 5.1 Si 2.46 12.4 Ca 22 27.3 Ti 40.1 47.3 Fe 115.6 128 Table 1. Path lengths of Selected Elements in Air and SiO2
Traditional sample preparation methods are fused beads and pressed pellets. Fused beads ensure the best homogeneity because the sample is crushed and the powder mixed with a flux, which is then heated to around 1000ºC and cooled to form a bead. Pressed pellets are made by using high pressure to flatten the powder into a little cake, either with or without a binder to hold it together. Pressed pellets require less work to prepare, but may have more issues with homogeneity than the fused beads. Sample preparation and lab analysis take time and the high level of accuracy required by ASTM C-114 may not be necessary in all parts of the cement-making process. The requirements in ASTM C-114 only govern the final product. Less accurate results can still be useful in homogenizing feed material or in other raw material decisions. What if less accurate but good enough results could be obtained with little or no sample preparation? This paper presents a study done with handheld XRF using powders, pressed pellets, and unprepared rocks to demonstrate the ability of handheld XRF to be used as a cost-saving device in the cement-making industry. EXPERIMENTAL Calibrations were developed for the Bruker S1 TURBO SD LE using pressed pellets and powders and were tested on a variety of relevant samples. The calibration made from powders was done by measuring known customer standards and certified geological standards. Powders were used as-is, with no additional grinding. They were placed in a sample cup covered with 4μm Ultralene foil and measured for 180 seconds. These data were used to create a fundamental parameters-based, semi-empirical calibration that reports MgO, Al2O3, SiO2, SO3, K2O, CaCO3, TiO2, MnO, and Fe2O3. This is the standard composition of limestone, the highest quality raw material for the cement-making process. The assumptions about the stoichiometry of each element are key parameters in obtaining accurate results. Based on the results obtained, two more versions of the calibration were created, using different stoichiometric assumptions. The first assumes Ca and Mg exist as carbonates and the rest of the elements are their standard oxides. This mimics the composition of dolomite. The second assumes Ca is also a standard oxide (CaO) along with the rest. This is the composition of ignited limestone/dolomite, including clinker and everything after the kiln. The calibration made from pressed pellets was designed for measuring unprepared rock surfaces. Pressed pellets more closely resemble rocks because there are not additional errors arising from packing density variations in the powder. The pellets were measured for 180 seconds each, and a semi-empirical calibration similar to that
with the powders was created. Three versions of this calibration were made, each using a different set of assumptions, to demonstrate the effect that stoichiometric assumptions have on the final performance of the calibration. The first version assumes Ca as a carbonate (CaCO3) and the other elements as their standard oxides. It also normalizes the results to 100% by summing the measured values and multiplying them by a factor to cause the sum to be 100%. This is the same set of assumptions used in the powder calibration. The second version also assumes the same stoichiometry but does not normalize the results. The third version assumes all elements to be in their standard oxide form (Ca = CaO) and does not normalize, but uses a CO2 matrix balance, which assumes all remaining percent composition (all unmeasurable material) is carbon dioxide. The matrix balance sometimes causes instabilities in the calculations, but still can often be helpful. Normalization is a very useful tool, but should be used carefully. It can correct for problems like large grain size variation or the sample not sitting flat on the nose of the instrument. Figure 1 shows the effect of normalization on the Ca content reported for measurements of the same sample with grain sizes varying from fine powder to larger rocks about the size of almonds. The normalized data only show about a two percent variation of the CaCO3 Figure 1. Normalization values, while the non-normalized data show around a 30% variation in the reported CaO. Note that the normalized calibration reports CaCO3 instead of CaO. This is because when using normalization, all the elements present must be accounted for, including the x-ray-invisible elements like C, O and H. If they cannot be measured, they must be included through stoichiometry. The absorption characteristics of the invisible elements are also used in the calculation of concentration from intensity for the measured elements. All test samples were measured for 180 seconds. Powder samples were in a sample cup with 4μm Ultralene foil. Each powder sample was shaken to homogenize, then tapped on a surface to ensure equivalent packing density in all samples.
RESULTS Powder Calibration The original calibration made from powder standards was tested on certified geologic powder standards. Table 2 contains the data for these tests. There is very good agreement between known and measured values on the limestone samples. As the stoichiometry diverges from the assumptions, the agreement in the data goes down. The dolomite and gypsum show deviations in the CaCO3 values because of their GeoMajors Geological Standards MgO Al2O3 SiO2 SO3 K2O CaCO3 TiO2 Mn2O3 Fe2O3 Limestone_04 known 0.15 0.12 0.70 0.02 0.02 98.8 0.009 0.010 0.045 measured 1.65 0.61 0 0.088 0.045 97.5 0.006 0.017 0.069 Limestone_14 known 1.94 0.405 3.04 0.118 0.030 87.7 0.049 0.178 2.715 Dolomite_05 Gypsum_19 SuperPhosphate_20 Clay Cement_01 measured 0.804 0.429 4.17 0.064 0.043 90.6 0.023 0.155 2.270 known 21.4 0.054 0.376 0.010 0.020 54.3 0 0.004 0.030 measured 20.1 0 0 0.032 0.114 79.6 0.004 0.010 0.075 known 1.74 0.34 1.68 51.9 0.095 70.1 0.019 0 0.153 measured 3.23 0 2.24 43.6 0.080 50.7 0.008 0.025 0.125 known 0.21 1.1 2.92 42.8 0 31.5(CaO) 0 0 0.4 measured 6.03 0.804 3.68 33.1 0.005 40.4 0.023 0.020 0.308 known 0.3 31.6 51.1 0.250 2.23 0.304 1.08 0 1.04 measured 0 25.8 70.0 0.009 1.91 0 1.02 0.01 1.24 known 0.42 4.85 21.8 2.25 0.110 70(CaO) 0 0 0.3 measured 1.27 2.38 22.8 1.62 0.078 71.6 0.008 0.030 0.178 Table 2. Powder Calibration, Geological Standards differing chemistries. Although the clay has good agreement in Ca, it diverges in SiO2 and Al2O3. This has not been fully investigated, but it is suspected that the divergence is caused by normalizing the calibration without including the proper loss on ignition (LOI). In clays, CaO is more likely to be present instead of CaCO3. Using an assumption of CaCO3 is not an accurate representation of the LOI for a clay-type material. This would cause deviations in the major elements, with the SiO2 divergence being the largest because it is the most prevalent element.
The superphosphate was chosen as a test sample to demonstrate the effect of choosing totally wrong stoichiometry. Because the stoichiometry assumed in the calibration is nothing like that of the superphosphate, and the P2O5 is not taken into account, rather wild results can be seen. This demonstrates the importance of at least having some idea of the mineralogy of a sample in order to get accurate results. If just a rough idea of the composition is needed, a generalized mining or soil cal can be used to get estimates of the elements present, even for unusual stoichiometries. An interesting sample was the cement sample, which is a final-product cement. Because this has been ignited in the kiln, the CO2 burned off, leaving CaO instead of CaCO3. The reported value for the calcium was correct, but the units are very misleading. A better assumption for the final cement product, as well as clinker and anything that has been previously ignited, is CaO. GeoMajors Geological Standards MgCO3 Al2O3 SiO2 SO3 K2O CaCO3 TiO2 Mn2O3 Fe2O3 Dolomite_05 known 44.8 0.054 0.376 0.010 0.020 54.3 0 0.004 0.030 Dolomite Assumptions measured 40.5 0 0 0.037 0.083 59.3 0.003 0.009 0.049 GeoMajors Geological Standards MgO Al2O3 SiO2 SO3 K2O CaO TiO2 Mn2O3 Fe2O3 Cement known 0.42 4.85 21.8 2.25 0.110 70 0 0 0.3 Cement Assumptions measured 0.565 3.6 22.1 1.850 0.12 71.4 0.01 0.04 0.37 Table 3. Dolomite and Cement Assumptions In order to test the hypothesis that stoichiometric assumptions were responsible for the deviations, a set of assumptions was applied to the calibration for dolomite. Another set of assumptions for gypsum were applied and the samples recalculated with these changes. Table 3 shows the results for dolomite (Mg-Ca-carbonate) and ignited cement (Ca oxide, no CO2). It is obvious that the assumptions were the cause of the deviations, and with the proper assumptions, good values can be obtained for these matrices. One thing to keep in mind when using the cement (CaO) assumptions is that if it is used to look at final product, additional elements could be present from blending with other materials. These would not be taken into account in this calibration, which is designed primarily for raw materials. Some customer samples were also tested. These samples are representative of the types of raw materials used in their process. Table 4 shows results for these. The same trends were seen as in the geologic standards. The limestones have very good agreement, with the clays diverging a little. However, all the materials tested showed good enough agreement to provide useful results. Decisions about raw materials can be made based
on this quality of results, thus saving time and money on sample preparation and lab analysis. Customer Samples MgO Al2O3 SiO2 SO3 K2O CaCO3 TiO2 Fe2O3 limestone C15A04e2bb known 0.63 0.3 1.28 0.02 0.08 96.8 0.02 0.28 measured 1.4 0.73 0 0.080 0.056 97.4 0.0063 0.316 precrusher stockpile C15B02f285 known 0.12 18.2 67.8 0.06 0.15 0.125 1.1 7.36 measured 0 17.2 72.6 0.002 0.22 0.147 1.1 8.6 stockpile C15B02f1c0 known 1.09 1.61 12.5 0.04 0.34 81.8 0.09 1.09 measured 0.48 1.18 13.1 0.083 0.29 83.5 0.033 1.26 feed limestone C15B02d59b known 0.67 0.76 4.72 0.42 0.18 91.1 0.05 0.99 measured 0 0.51 5.69 0.52 0.18 91.8 0.027 1.19 cement fringes C15B02d348 known 0.28 1.22 5.27 41.0 0.22 53.2 0.04 0.59 measured 4.15 1.16 6.58 37.6 0.26 49.5 0.042 0.69 weekly clay C15B02a6cf known 0.2 28.0 57.2 0.23 0.39 0.77 1.36 2.91 measured 0 28.6 65.2 0.029 0.44 0.94 1.57 3.25 clay-birdwood C15A04e2bc known 0.29 23.7 66.8 0.04 1.03 0.09 1.08 0.7 measured 0 27.6 69.2 0.0022 1.09 0 1.22 0.90 Table 4. Powder Calibration, Cement Process Samples Pressed Pellet Calibration Pressed pellets were chosen for the calibration to be used for unprepared rocks in order to more closely simulate the rocks than a powder calibration. This is because the pellets don't have added uncertainties due to variations in the packing density of the powder. As seen in the powders, choosing the stoichiometry closest to that of the samples to be measured is very important because the absorption characteristics of the "invisible" elements are used in the calculation. For high quality raw materials used in cement manufacturing, the CaCO3 assumption is usually valid. If other compositions are known to be prevalent, the assumptions of a calculation can be easily changed to accommodate the new stoichiometry.
Table 5 shows data taken on unprepared rocks, as well as on pressed pellets made from those rock samples. For each sample number, a bag of rock chips was provided. From this bag, three representative samples were chosen for measurement. When there was a visible color difference between the rocks in a bag, samples were chosen to represent the whole range of colors. Known CaCO3 No Norm (CaCO3) Table 5. Rock and Pellet Data Known CaO CO2 Matrix (CaO) Norm (CaCO3) Sample # 1 pellet 96.3 98 98.4 53.93 55.8 raw ave 96.7 95 51.3 raw range 95.8-97.9 91.7-98.1 43.6-57.9 2 pellet 89.48 91.6 92.1 50.11 52.4 raw ave 89.5 87.8 47.6 raw range 77.4-96.2 71-97.8 32.5-59.9 3 pellet 74.2 80.7 78.1 41.56 40.5 raw ave 79.7 75.3 37.5 raw range 78.7-81.3 72.4-80.4 31.4-46.8 4 pellet 88.14 88.1 90.3 49.36 54.3 raw ave 89.9 96.1 67.4 raw range 88.1-92.9 93.6-98.7 59.3-79.5 5 pellet 81.43 85.6 85.1 45.6 46.9 raw ave 86.8 85.4 47.2 raw range 76.5-92.1 68.5-94.2 30.2-56.4 6 pellet 79.8 84.3 83.5 44.69 45.7 raw ave 79.6 81.5 47 raw range 78.2-84.4 80.4-82.7 43.4-49.1 7 pellet 97.25 97.3 99.3 54.46 59.1 raw ave 96.2 100 71.2 raw range 94.3-99.1 100-100 58.6-77.8 8 pellet 85.43 85.6 87.4 47.84 51.9 raw ave 69.5 75.7 58.2 raw range 32.2-89.8 32.9-99.4 19.3-92.8 9 pellet 75.78 80.7 79.3 42.44 42.7 raw ave 81 77.6 39.3 raw range 80.1-81.9 76.4-79.4 38.3-41.2 The data shows that although there can be a large range in measured values for the rocks, the average value is generally close to the known value. This demonstrates the importance of taking multiple measurements on inhomogeneous samples. The
normalized calibration showed the least variation in the rock samples, but not by a huge amount. The non-normalized CaCO3 had slightly higher variation, but approximately the same average deviation from the known. The CaO with CO2 matrix had the most variation in the rock samples and deviated significantly on more samples than the others. One interesting anomaly is Sample #8. This sample was visibly heterogeneous within each rock chip, with spots of white (presumably SiO2) on a grey background. This sample showed a huge deviation between measurements, and the average was not that close to the known value. This is a case when taking more than three measurements is important. If five or ten measurements were taken, it is likely that the average value would approach the known value. When there is visible heterogeneity, it is very important to take multiple measurements. CONCLUSION These data suggest that the use of handheld XRF in measuring raw materials in the cement-making process may result in significant time and money savings. The data on powders and even on unprepared rock samples show that good enough results can be obtained to make decisions in the field. While handheld XRF will not replace the benchtop or laboratory instruments in the cement industry, it can provide a cost-saving enhancement to the users that will enable them to make faster decisions about raw material management.