Testing of industrial products - Aggregates for construction

Details of the cross-testing experiments on the Schlagversuch

Laboratories

Sixteen laboratories in Germany took part in the experiment. The laboratories have been given numerical codes. For the purposes of this report they have also been assigned letter-codes (because single-character codes are required for some of the graphs).

Materials

Laboratory samples of three aggregates were prepared by Partner 1 and distributed by Partner 3. The samples were prepared, for each level of the experiment, as if they were laboratory samples taken from one bulk sample. A report describing their preparation is available (Delalande, 1994. Strength tests programme; preparation of laboratory samples. LCPC, Paris, 24 November, 1994.). The same three aggregates were used throughout the four-year programme for all the cross-testing experiments that involved tests of the mechanical properties of aggregates. For the Schlagversuch, laboratory samples of two size fractions (6/10 mm and 10/14 mm) of each aggregate were sent to the participants

The participating laboratories were required to prepare six test specimens from each aggregate, according to the draft CEN method used in the experiment. This involved combining the samples of the 6/10 mm and 10/14 mm fractions of an aggregate to form a "laboratory sample", sieving this laboratory sample into three size fractions (8/10 mm, 10/11.2 mm and 11.2/12.5 mm), then dividing the fractions and recombining them to produce specimens of the required mass.

The participating laboratories were also required, as part of the test method, to determine the particle density of each aggregate and use their values to calculate the specimen mass to use in their tests.

The methods used to prepare the laboratory samples, and test specimens from laboratory samples, allow the precision of the test method to be described using Wc , r1 and R1 as used here.

Data

The Schlagversuch method requires single determinations for individual test specimens to be rounded to 0.01 %, and the test results (calculated as the averages of determinations on three specimens) to be rounded to 0.1 %. However, to ensure that the estimates of repeatability and reproducibility standard deviations are not affected by rounding of the data, the weights retained on the test sieves, as reported by the participating laboratories, were used to re-calculate individual determinations and SZ values, and these values were used to calculate the standard deviations without any rounding (other than that which occurs because of the limited precision of the computer used). In the data tables later in this report the determinations and SZ values are rounded to the nearest 0.01 %.

Averages and ranges

The averages and ranges are shown in the histograms, and the averages are shown in the Mandel plots.

Between-specimen standard deviations are used to calculate the critical range, and between-specimen ranges or standard deviations to assess the repeatability of tests from individual laboratories. Between-test-result ranges are used to calculate the repeatability of the test method, and to assess the repeatability of tests from individual laboratories also. Laboratory averages are used to calculate the reproducibility of the test method, and to assess laboratory biasses.

Grubbs' test and Cochran's test have been used to identify stragglers and outliers in the data. Stragglers are marked in the tables and figures with "?", and outliers with "??". The procedure for applying these tests is described in ISO 5725-2, 1994. The data in this experiment contain a factor, "specimens", not considered in ISO 5725-2, and so the ISO procedure has to be extended by applying Cochran's test first to the between-specimen standard deviations and then to the between-test-result ranges. Finally Grubbs' test is applied to the laboratory averages.

The strength data for Laboratory G (194) contained stragglers and outliers in Levels 1 and 2, and also their density result for Level 1 was high in comparison with those reported by the other laboratories. They found that they did not have enough material in the samples that they had been sent originally for the test, and had to be sent additional samples of the materials for Levels 1 and 2. Their data for Levels 1 and 2 were excluded from the precision calculations. Likewise Laboratory J (202) experienced similar difficulties and their data for Level 1 gave an outlying laboratory average, so these data were excluded from the precision calculations.

The histograms and the X-Y plots and plots of standardised biasses, standard deviations and ranges can be used to identify laboratories that give consistently (i.e. over more than one level of the experiment) high laboratory biasses or ranges.

Sensitivity ratios.

See Table 4 of the "results". It is of interest to compare the precision of the Schlagversuch with that of other tests for measuring the mechanical properties of aggregates, such as the Los Angeles test. However, the standard deviation of repeatability for the Schlagversuch cannot be compared directly with the standard deviation of repeatability of other mechanical tests, because of differences in the units of measurement, and differences in their scale. The same drawbacks apply to the direct comparison of the standard deviation of reproducibility for the Schlagversuch with the standard deviation of reproducibility of other mechanical tests.

A comparison of the precision of the Schlagversuch with the precision of other mechanical tests is made possible by the comparison of dimensionless "sensitivity ratios". From the formulae, it will be seen that the sensitivity ratio uses the difference between the average results reported for two aggregates. Thus it is essential that sensitivity ratios for different test methods are derived from tests on the same aggregates if they are to provide a valid basis for comparison of the test methods.