The Chemical Analysis of Argonne Premium Coal Samples

Edited by Curtis A. Palmer
U.S. Geological Survey Bulletin 2144

Compilation of Multitechnique Determinations of 51 Elements in 8 Argonne Premium Coal Samples

By Curtis A. Palmer and Sarah A. Klizas

ABSTRACT

Eight Argonne Premium Coal samples were analyzed by the U.S. Geological Survey. The concentrations of 51 elements were determined by two or more techniques on each sample. The analyses were performed by energy- and wavelength-dispersive X-ray fluorescence spectrometry, instrumental neutron activation analysis, inductively coupled argon plasma-atomic emission spectroscopy, atomic absorption spectrometry, inductively coupled argon plasma-mass spectrometry, and direct-current arc spectrographic analysis. All data are compiled on a whole-coal basis for ease of comparison. The ash values are also included so that data can be converted to an ash basis if desired.

INTRODUCTION

Although the eight Argonne Premium Coal samples analyzed in this study are not defined as 'reference standards' by Argonne National Laboratories, they are extremely important because of the care that has been taken in collection, preparation, and storage. A detailed description of the background information for these samples has been reported by Vorres (1990, 1993). However, these samples have not been widely analyzed for trace elements. The analytical laboratories of the U.S. Geological Survey analyzed these samples to further characterize them and to provide a foundation for a trace-element data base.

Most quantitative techniques used for elemental analyses of geologic samples offer high levels of precision and accuracy for selected elements in certain types of samples over specific ranges of concentrations, but all analytical techniques have certain characteristic limitations. For example, matrix-induced spectral interferences can result in incorrect determinations of trace elements. Even if properly corrected, these interferences may lead to reduced sensitivity or precision for a given element. Generally, the concentrations of elements determined by another technique on the same matrix will not be affected by the same interferences.

A multitechnique approach for major- and trace-element analysis was taken to provide the high degree of reliability desired to characterize these materials. In addition, this information may be useful in evaluating data from a single technique for coal analysis for laboratories that do not have all techniques available. Semiquantitative analytical techniques, although not offering the precision or accuracy of the quantitative techniques, rapidly provide a large volume of data. Some of the data obtained by these low-precision techniques are not easily obtained by quantitative methods, but can be useful in the overall characterization of these materials.

This paper

1. summarizes the results of the multitechnique analyses of the Argonne Premium Coals,
2. discusses some discrepancies in the data, and
3. determines 'recommended values' or 'best averages' depending on the precision of the data.

Each of the eight Premium Coal samples has been analyzed in triplicate for 68 elements. Fifty-one elements were determined by more than one technique. Although up to seven different techniques were used for some elements, there are not enough high-precision data to recommend values for all elements in all coals using common criteria for establishing such values (Kane and others,1990). Therefore, modified criteria were designed for this data set. They allowed definition of 'recommended values' on slightly less than half of the elements included in the data set.

SAMPLES AND TECHNIQUES

Three splits of each of the Argonne Premium Coal samples were analyzed by multiple techniques. The samples and the sample identification protocol are described in this volume by Palmer.

Ideally, solid samples of the whole coal would be analyzed by instrumental techniques because this type of analysis avoids problems caused by volatilization of elements during ashing and problems caused by incomplete sample dissolution. The procedures used for determining element concentrations instrumentally on the whole coal are discussed in this volume in the following papers:

Although the sensitivity of INAA was acceptable for most of the 29 elements determined, the sensitivities of the other whole-coal procedures were marginal for many elements. Therefore, coal ash procedures were also used for WDXRF, EDXRF, and DCAES (see list below) to concentrate the trace elements and thereby increase sensitivities. Techniques that require analysis of coal ash were used as described in the following papers in this volume:

All samples were ashed at 525°C to limit volatilization of lead, cadmium, and other moderately volatile trace elements. Ash yields were determined on the same splits used for the analyses and were used to calculate data as if determined on a whole-coal basis. The 525°C ash yields, which are not directly comparable to those determined by ASTM (American Society for Testing and Materials, 1996) ash procedures (750°C) but are generally similar, can be used to recalculate back to an ash basis if desired.

DISCUSSION OF RESULTS

Direct comparison of results presented in previous papers in this volume is difficult because data are presented in three different forms depending on the analytical technique used and the material analyzed. Concentrations are reported on an ash basis for some procedures, on a whole-coal basis for some procedures, and on an oxide basis of the ash for major elements determined by WDXRF and DCAES. To facilitate a direct comparison of the data, the ash data have been recalculated to whole-coal values and converted to an element basis for those elements reported on an oxide basis. The entire recalculated data set for all splits can be found in appendix 1. The number of significant figures given in the original papers has been maintained in the converted values.

A careful examination of appendix 1 shows that analytical procedures can be classified into two categories: highly precise (HP) procedures shown in bold, which generally have a relative standard deviation of less than 5 percent, and procedures that are less precise (LP). The precision was calculated by determining the percent of difference between the three individual data points and their mean for each sample-element pair. Using the accuracy guidelines discussed later in this paper, each test for each element was given a rating of good, usable, or poor precision. Comparisons of the different ratings for all elements determined by each technique were made. Finally, the techniques were divided into the two precision groups (LP and HP) based on which rating they received most frequently. In this study, the two DCAES procedures (ash and whole coal) and the X-ray whole-coal procedures were classed as LP procedures; INAA, ICAP-AES, ICAP-MS, CVAAS, HGAAS, FAAS, GFAAS, and the other X-ray procedures were classified as HP procedures. It should be noted that no procedures had the same precision for all elements in all samples. For the designated HP techniques, most determinations were of high precision, but as expected, determinations near the detection limit for some samples had poorer precision. LP procedures generally had lower precision for all samples and elements.

Statistical approaches are useful for large data sets; however, often they do not provide the detail that is useful in evaluating individual problems in the data. Even though the individual samples were analyzed only in triplicate, the complete data set requires 18 pages (appendix 1). A summary of the data is given in tables 1 and 2. Table 1 presents the method averages of the major rock-forming-element data determined on each of the three splits of the eight Argonne Premium Coal samples. Table 2 is a similar table for the trace-element data.

Statistical analysis of the data in appendix 1 is given in appendix 2. These data include the number of samples for which values were determined, the arithmetic mean (mean), the standard deviation, the relative standard deviation, the geometric mean, the deviation of the arithmetic and geometric means, and an analysis of the kurtosis and skewness for the HP techniques excluding outliers (values with only one significant figure and HP values excluded because of the 40 percent rule discussed in the next paragraph). A similar analysis for all values, including LP values, outlier values, and other excluded values, is also given in appendix 2.

Another approach to analyzing the data is to define the agreement between techniques in a useful, nonstatistical manner and then discuss individual cases of disagreement. In a practical sense, for major elements (elements with concentrations generally greater than 0.1 percent; table 1), procedures are said to have 'good accuracy' if the standard deviation of the individual determinations for a given sample determined by a given technique is ±5 percent of the mean of all of the HP procedures and does not disagree by more than ±0.5 percent absolute. For trace elements (table 2) 'good agreement' is defined as ±10 percent of the mean. 'Usable agreement' is four times the uncertainty of 'good agreement' or ±20 percent for the majors and ±40 percent for the traces. Excluded from the agreement analysis were values of only one significant figure. If more than two tests were used, a mean value for a given technique differing by more than 40 percent from the mean of the remaining values was reason for excluding a given technique. This is the '40 percent rule.' The excluded technique was said to have poor agreement for elements in those samples. In addition, the Grubbs test (Taylor, 1987) was made for all suspected HP outliers, using the mean and standard deviation for all HP techniques. Outliers are reported in table 3 under exceptions. They were excluded from the determination of agreement except for cases where the outlier was the only value for an element determined by a given technique. Figure 1 summarizes the decisions required to determine the agreement.

Trace-element criteria were applied to those samples containing elements that are traditionally considered major or minor elements (table 1), but whose mean concentration for HP techniques was less than 0.1 percent. This included phosphorus and manganese for all samples and magnesium, sodium, potassium, and titanium (see table 3) for four or more samples. It should be noted that the ±5 percent criterion for 'good agreement' is better than expected for some HP techniques for some samples as concentrations approach the detection limit. For example, counting errors of as high as 28 percent are reported for potassium by INAA (see paper by Palmer, this volume). The criterion of 20 percent required for 'usable agreement' is much smaller than the inherent precision for many of the LP techniques. Skeen and others, in this volume, report possible errors of +50 percent or -33 percent because of the nature of the standards for each of the elements. It is not surprising, therefore, that not all HP techniques have 'good agreement' for all samples and that most LP techniques have 'poor agreement' for most samples. Most of these disagreements were within ±50 percent of the HP mean concentrations, and scatter in inter-technique comparison plots simply demonstrates the poorer precision of the LP techniques. A complete discussion of the precision of all values in this study is beyond the scope of this paper, but an indication of the precision can be obtained by examining appendix 1 and by the relative standard deviation given in appendix 2.

A summary of the agreement for major elements is given in table 3. Data for elements by specific techniques were classified as in 'overall good agreement' with the mean of the HP procedures if at least half of the individual samples were in 'good agreement' using the previously mentioned criteria, and no samples had 'poor agreement.' If half or more of the samples determined were in 'good' or 'usable agreement' and the technique was not classified as in 'good agreement,' the element had 'overall usable agreement.' All others had 'overall poor agreement' except where the technique was the only HP technique. In this case, agreement could not be determined. The mean of the samples for this sole HP technique was used to assess the agreement of the LP techniques, and no accuracy designation was given. All HP techniques listed in table 3 were in 'overall good agreement' except for sodium, magnesium, manganese, and phosphorus determined by WDXRF on the ash, and potassium determined by ICAP-AES in Reston. These were classified as having 'overall usable agreement.' In contrast, all LP techniques had 'overall poor' or 'overall usable agreement.'

Table 4 summarizes the agreement of trace elements. All agreements were evaluated using trace-element criteria discussed previously and summary classifications similar to those in table 3. HP procedures generally had 'good agreement' and none had 'poor agreement,' whereas LP procedures generally had 'usable' or 'poor agreement,' with only an occasional 'good agreement.' Most of the data fall within expected precision limits, but barium is an example of a case where determining an element by more than one technique can make a significant difference. Agreement between techniques can be graphically represented by plotting the concentrations of all elements determined by one technique versus all corresponding concentrations by a second technique and comparing these points to a theoretical line with zero intercept and a slope of 1. Figure 2, for example, shows the comparison of INAA and ICAP-AES data from table 2 for all elements that the two techniques have in common. There is relatively little scatter (excellent agreement) in most of the data; therefore, the few problems with the data are easily recognizable. The most obvious discrepancy in the data is that the barium concentration determined by ICAP-AES is more than an order of magnitude smaller in WY PC-2 and POC PC-5 than the concentration determined by INAA.

The data for barium determined by the seven different techniques are shown in figure 3, plotting barium determined by all techniques versus the mean barium concentration determined by high-precision techniques. Although there is scatter among data from different techniques, ICAP-AES (R) data for WY PC-2 and POC PC-5 are clearly off the correlation line. The disagreement of ICAP- AES (R) data with data from all other techniques suggests that barium is present in a species, probably BaSO₄ , which is not dissolved by the Reston ICAP-AES acid dissolution procedures (see paper by Doughten, this volume). ICAP- AES procedures done in Denver use a sinter dissolution procedure and yield Ba concentrations that agree with the INAA data (see paper by Briggs, this volume). Both of these coals contain enough sulfate sulfur (Vorres, 1990) to account for all barium being BaSO₄ in the original coal. Solubility studies of these coals by Finkelman and others (1990), however, show that barium in these two samples is soluble in ammonium acetate and is therefore readily exchangeable. This suggests that BaSO4 is not in the original samples of WY PC-2 and POC PC-5 but that it is formed in the ashing process. Clearly, the stronger dissolution procedures (see paper by Briggs, this volume) should be used.

Some interesting consequences result from the 'agreement' rules. Although zinc determined by EDXRF on the ash had 'good agreement' in six samples, the mean of the zinc values for IL PC-3 was only slightly greater than one half of the zinc values determined by INAA and ICAP-AES for that sample. This difference led to a classification of 'poor agreement' for zinc in IL PC-3 and an overall 'usable agreement.' Although these results are unusual and can be treated as outliers, as will be discussed later, they relate to the overall reliability of a technique. Outliers also led to usable ratings for copper determined by ICAP-AES (Denver), arsenic determined by ICAP-AES (Denver), yttrium determined by EDXRF, and barium determined by ICAP-AES (Reston). Barium had only 'overall usable agreement' because of the 'poor agreement' in WY PC-2 and POC PC-5, probably caused by incomplete dissolution. Strontium had one outlier ('poor'), three 'good,' and four 'usable' values.

All techniques for cadmium, lanthanum, samarium, and uranium showed only overall 'usable agreement' even though some samples showed 'good agreement' for each of these elements. In these cases it is difficult to determine which technique may be in error. For cadmium, samarium, and uranium, there are only two HP techniques. For years INAA has been considered an excellent technique for the rare earth elements and may provide the best data for lanthanum and samarium. EDXRF is generally not considered to be the best technique for rare earth elements such as lanthanum because of spectral overlaps and values generally near the detection limits. ICAP-MS should produce good results for the rare earth elements but does not always agree well with INAA. This disagreement may be due to incomplete dissolution of some rare-earth-bearing species such as zircon.

Not surprisingly, EDXRF values for cerium yielded only 'usable agreement' because cerium concentrations were very near the detection limits for this technique. Other techniques for cerium showed 'good agreement.' However, chromium determined by EDXRF, well above the detection limit, also showed only 'usable agreement,' whereas all other techniques showed 'good agreement.'

The overall agreement of niobium determined by EDXRF and neodymium determined by INAA was usable only because several samples had values at or below the detection limits. Ge, Ga, Mo, Ag, Sn, Pr, and Bi all had only one HP technique, so no rating could be determined Beryllium and vanadium were determined only by ICAP-AES in both Reston and Denver. The 'good agreement' that was expected for these techniques is printed in italics in table 4 because these techniques are modifications of the same technique.

The causes of all of the discrepancies are not known. Overall, however, the data are generally useful and provide an excellent base for further study.

DETERMINATION OF RECOMMENDED VALUES

Because some elements were determined by only one high-precision technique, because some element concentrations approached their detection limits, and because some samples contained interfering elements, recommended values cannot be reliably calculated for all elements in all coals. Tables 5 and 6 present recommended and average values for concentrations determined by high-precision techniques.

Recommended values were determined by using procedures similar to those used in determining agreement ratings. The mean value was considered a recommended value if the relative standard deviation of all individual determinations of HP techniques excluding outliers was less than 5 percent for major elements or 10 percent for trace elements using the criteria discussed earlier, and there were at least four individual determinations. If a recommended value could not be determined, the value for the statistical parameter responsible for rejection was boxed in appendix 2, and an average of all HP techniques was reported in parentheses in Tables 5 and 6. Values excluded from the determination of recommended values and the reason for exclusion are given in the final column of Tables 5 and 6. After the analyses were completed, 43 percent of the values reported in Tables 5 and 6 were recommended values.

CONCLUSIONS

A multitechnique approach is the best method to differentiate 'good' values from 'poor' values. Differences are caused by spectral interferences, volatilization due to ashing, or incomplete sample dissolution. Interferences for a given element usually differ for each technique. Losses caused by volatilization can be determined by comparing data from whole-coal procedures and ash procedures. Insolubility problems can be identified by comparing data from techniques not requiring dissolution with data from techniques requiring dissolution.

This paper does not recommend values for all elements, but it does provide reliable data for many trace elements. It provides manipulations of the data that will allow readers to make their own interpretations and judgments. It also demonstrates that some techniques are more reliable than others for individual elements, and they depend on the concentration of an element. This paper shows that the more high-precision tests that can be run on a sample to measure certain elements, the greater the reliability the data and the greater the likelihood of determining a recommended value. More important, it points out the uncertainties in attempting to obtain reliable data from a single technique for coals of widely differing types, it provides a basis for determining some uncertainties of the techniques, and it should aid in the evaluation of data determined by different techniques.

REFERENCES

American Society for Testing and Materials (ASTM), 1996, ASTM Designation D 3174-93, Standard test method for ash in the analysis sample of coal and coke from coal: 1996 Annual Book of ASTM Standards, v. 05.05, Gaseous fuels; Coal and coke, p. 291-294.

Finkelman, R.B., Palmer, C.A., Krasnow, M.R., Aruscavage, P.J., Sellers, G.A., and Dulong, F.T., 1990, Combustion and leaching behavior of elements in the Argonne Premium Coal samples: Energy and Fuels, v. 4, no. 6, p. 755-767.

Kane, J.S., Arbogast, B.F., and Leventhal, J.S., 1990, Characterization of Devonian Ohio Shale SDO-1 and a geochemical reference sample: Geostandards Newsletter, v. 14, p. 169-196.

Taylor, J.K., 1987, Quality assurance of chemical measurements: Chelsea, Mich., Lewis Publishers, Inc., 328 p.

Vorres, K.S., 1990, The Argonne Premium Coal Sample Program: Energy and Fuels, v. 4, no. 5, p. 420-426.

------1993, Users handbook for the Argonne Premium Coal Sample Program: Argonne National Laboratories Report ANL/PCSP-93/1, 200 p.

[ Contents ]

Created by the EERT WWW Staff.