Estimates of repeatability and reproducibility are given in Table 1. The standard deviations are also shown in Figure 1.
Note that these standard deviations (and the values given for r, r1, R and R1) are estimates that can be influenced by individual results. Thus in Level 1 the data for Laboratories B and I gave some stragglers in the between-test-portion ranges, and the data for Laboratory C give a straggler in the laboratory averages. In Level 3 there are no stragglers but the data for Laboratories B and L gave unusual laboratory averages. If the data that give rise to these values are not included in the calculations, the repeatability standard deviation for Level 1 would be reduced from 0.012 to 0.009 and that for Level 3 would be reduced from 0.123 to 0.096. The reproducibility standard deviation for Level 1 would be reduced from 0.014 to 0.010, and that for Level 3 from 0.537 to 0.326. However, it is doubtful if estimates obtained by excluded such a large proportion of the data give a valid measure of the precision of the test method.
The functional relationships given in Table 2, and shown as the straight lines on Figure 1, were calculated using the standard deviations for Levels 1 and 3. Data for only two materials are not sufficient to establish these relationships with any degree of confidence.
Figure 2 shows results that are given in the current British Standard method for the determination of acid-soluble sulfate (BS 812 Testing aggregates; Part 118 Methods for the determination of sulfate content. British Standards Institution.). These were obtained from an experiment involving five aggregates (in the order of their sulfate contents: a 14mm limestone, a 14mm sea-dredged gravel and sand, a 5mm crushed limestone/shell mixture, a 14mm blastfurnace slag, and a 28mm colliery shale). The design of the British experiment was similar to the cross-testing experiment reported here. Samples of aggregates (not powdered materials) were distributed. The sample reduction procedure in the British Standard requires similar masses to be produced as the material is crushed and ground as in the draft European Standard on sample reduction (Proposed prEN 932-2 Tests for general properties of aggregates; Part 2 Methods for reducing laboratory samples to test portions. CEN/TC 154/TG 5 committee paper N 188. June, 1994.). The method of determination of acid-soluble sulfate in the British Standard is similar to that in the draft European Standard method. Details of the British experiment are given in the reference. It can be seen in Figure 2 that the precision estimates depend on the level of acid-soluble sulfate by functional relationships that are well-approximated by straight lines. Hence the results of the British experiment may be used to justify the fitting of straight lines to the results of the cross-testing experiment.
However, comparison of Figures 1 and 2 shows that, in general, lower values for the repeatability and reproducibility were obtained in the British experiment than in the recent European experiment. The discussion of stragglers and "unusual" results above suggests that the precision achieved in the European experiment has been affected adversely by these results, and it may be that laboratories could achieve better precision if they had more experience with the method.
The proposed draft European Standard specification for aggregates for concrete (Proposed draft for aggregates for concrete including those for use in roads and pavements. CEN/TC 154 committee paper N 232. January, 1994.) contains the following information:
"The acid-soluble sulfate content of the combined aggregates for concrete shall not exceed:
1.0% SO3 for slags and
0.2% SO3 for aggregates other than slags."
The materials used in the cross-testing experiment gave overall average sulfur contents of 0.06% in Level 1, 0.37% in Level 2, and 1.61% in Level 3. They thus cover the range of acid-soluble sulfate contents that will be of interest in practical situations when the European Standard comes into effect.
It has been argued (J»rck, Sym and Powell, 1994. A study of mechanical tests of aggregates. Green Land Reclamation Ltd Report GLR 3036/03a) that the reproducibility standard deviation of a mechanical test, when expressed as a coefficient of variation, should be no more than about 8%, if the test method is to be used to assess the compliance of aggregates with specifications. The argument used to justify this value applies equally-well to non-mechanical tests when their results are to be compared with upper limits in specifications, so it can be applied to the determination of acid-soluble sulfate contents.
The reproducibility coefficient of variation (see Table 3) for Level 2 (the powdered recycled concrete) is a little greater than 8%: this indicates that the reproducibility of the analytical method is not good enough for it to be used as the reference method for the determination of acid-soluble sulfates. It is interesting to note that in the recent cross-testing experiment involving the determination of total sulfur by the proposed European Standard reference method, in which powdered samples were also distributed (The proposed CEN method for the determination of total sulfur in aggregates: results of the 1994/5 cross-testing experiment. Green Land Reclamation Ltd Report GLR 3036/09.), similar reproducibility coefficients of variation were found (10.6%, 12.0% and 6.7%).
For Levels 1 and 3 (the concreting sand and the blast-furnace slag), where the participants were required to carry out the sample reduction themselves, the reproducibility coefficients of variation are much greater than 8%. This demonstrates that the sample reduction stage of the test method is a major source of between-laboratory variation, and renders the method unsuitable for use as a reference method.
If an aggregate (that is not a slag) has an acid-soluble sulfate content on the above specification limit of 0.2%, then the reproducibility standard deviation would have to be 0.016% SO3 for the reproducibility coefficient of variation to be 8%
(as 100 x 0.016 / 0.2 = 8.0%).
The result of the test is calculated as:
SO3 = 34.30 x m7 / m6 %
m6 = the mass of the specimen in grams
m7 = the mass of the barium sulfate precipitate in grams
This may be re-arranged as:
m7 = m6 x SO3 / 34.30
and used to calculate that a reproducibility standard deviation of 0.016% in the determination of SO3 is equivalent to a reproducibility standard deviation of 0.9mg in the determination of the mass of the precipitate, as:
2 x 0.016 / 34.30 = 0.0009g = 0.9mg
The test method requires masses to be determined to the nearest 0.1mg, so that the above precision is achievable in the weighings, but clearly great care needs to be taken if this precision is to be achieved when maturing, filtering and igniting the precipitate.
The reproducibility standard deviation obtained at Level 1 in the cross-testing experiment, on a sand that gave an average of 0.06% SO3, was 0.014% SO3 (see Table 1). This demonstrates that a reproducibility standard deviation as low as 0.016% SO3 is achievable in practice (although not necessarily achievable on an aggregate with a higher sulfate content). It is also interesting to note that at Level 1 the repeatability standard deviation was almost as large as the reproducibility standard deviation, showing that at such low sulfate contents, random errors in the analysis make a significant contribution to the overall variation. It seems to be making unnecessary demands on the skill of the analyst to start with the same specimen mass (2g) when testing an aggregate with an acid-soluble sulfate content at the 0.2% level as at the 1.0% level. If 2g is "right" for the 1.0% level, then 10g should be "right" for the 0.2% level: it is therefore proposed that the specimen mass should be increased to, say, 5g to 10g when testing for compliance with a limit of 0.2% acid-soluble SO3.
Many aggregates have acid-soluble sulfate contents well below the specification limit of 0.2%, and many slags have acid-soluble sulfate contents below their specification limit of 1.0%. For such materials, poor reproducibility of the determination of acid-soluble sulfate is not a problem. Thus if an aggregate has a true acid-soluble sulfate content less than 0.1% it is unlikely that a laboratory will report a result in excess of 0.2%. If a slag has a true acid-soluble sulfate content below 0.5% it is again unlikely that a laboratory will report a result in excess of 1.0%. For such materials it may be sufficient to determine total sulfur (by the relatively precise method of X-ray fluorescence spectrometry) and use the results to demonstrate that if all the sulfur is present as sulfate then the aggregate complies with its specification.
Poor reproducibility is a problem only for those materials that have acid-soluble sulfate contents in the neighbourhood (say, within three times the reproducibility standard deviation) of their specification limits: acceptable materials can then fail the test and unacceptable materials can be deemed to comply with the specification.
If the method is to be retained as the reference method, then there is a need to improve its reproducibility.
With regard to the analytical method, no other test methods exist for determining acid-soluble sulfate that are widely-available. There are a number of general methods for improving reproducibility that should be considered:
In addition, with materials that give results in the neighbourhood of their specification limit, it would be advisable to adopt a "belt and braces" approach: determine total sulfur, acid-soluble sulfide and acid-soluble sulfate, and check that the sulfide and sulfate add up to the total sulfur. A potential source of error in the determination of acid-soluble sulfate is oxidation of sulfide to sulfate: this check would reveal if this particular error had occurred. Methods are given in the proposed European Standard tests for chemical properties of aggregates (Draft prEN 1744-1 Tests for chemical properties of aggregates. Part 1 Chemical analysis. November, 1994.) for determining all three properties. Research would have to be carried out to establish the tolerance to be allowed on the difference
total sulfur - (sulfate + sulfide)
for the results to be deemed acceptable.
However, the results of the cross-testing experiment suggest that sample reduction is also an important cause of between-laboratory variation - probably more important than the analysis. The three general methods for improving reproducibility listed above could be used to improve the reproducibility of the sample reduction procedure. At present the test method does not specify the sample reduction procedure in any detail (the wording is such that it does not actually require that the proposed European Standard on sample reduction should be followed during the crushing and grinding operations). It is possible that the reproducibility would be improved if:
It should also be noted that the repeatability of the analytical method, as measured by the critical ranges (see Table 1), was much better at Level 2 (using the material ground by Partner 4) than at the other two levels. This suggests that there may be potential for improving the precision of the analysis by homogenising the ground material thoroughly.
Oxidation of sulfides to sulfates can occur during grinding, as well as during the analysis. Research has been done (Taylor, 1992. Blast furnace slag quantification in hardened concrete. DoE report PECD 7/6/185.) that shows that, in steel mills, the high temperatures that arise during grinding, in combination with iron filings worn from the working faces of the mill, can cause oxidation of sulfides. The "belt and braces" approach described above will not detect this problem because specimens for all three test methods are prepared by the same sample reduction process. It is therefore proposed that the reference method for the determination of acid-soluble sulfates should specify the use of grinding equipment with non-ferrous (e.g. silica, corundum or porcelain) working faces.
Other chemical tests for aggregates require the preparation of ground specimens (e.g. determination of total sulfur). The results of this experiment suggest the possibility that sample reduction could be a significant source of variation in these other methods: any research that is done on sample reduction should examine the benefits for all methods that require it.
The draft European Standard (Draft prEN 1744-1 Tests for chemical properties of aggregates. Part 1 Chemical analysis. November, 1994.) contains a general requirement that the results of chemical tests should be reported to the nearest 0.01%.
For the slag used in Level 3, the repeatability standard deviation in Table 1 is 0.123% SO3, and the critical range is 0.199 % SO3, so that rounding to the nearest 0.01 is needed to check repeatability. However, with the limit on acid-soluble sulfate that may be applied to slags, acid-soluble sulfate contents around 1% will be of interest in practice, so for the purpose of checking compliance of slags with specifications it would be more appropriate to report acid-soluble sulfate contents to the nearest 0.1%.
For the other aggregates used in Levels 1 and 2, the repeatability standard deviations are 0.012% and 0.004% SO3, and the critical ranges are 0.031% and 0.017% SO3, so that rounding to the nearest 0.001% is needed to check repeatability. With the limit on acid-soluble sulfate that may be applied to other aggregates, acid-soluble sulfate contents around 0.2% will be of interest in practice, so for the purpose of checking compliance of other aggregates with specifications it would be more appropriate to report acid-soluble sulfate contents to the nearest 0.01%.