Traditionally it has been assumed that construction material properties follow relatively normal patterns in data distributions. The assumption of normal distribution has worked well for such parameters as percent air voids and asphalt content in bituminous mixtures. In recent years, more agencies in the United States have been adopting statistically based specifications which recognize the variability inherent in materials as well as test methods. Many agencies are also considering the adoption of new testing procedures which are more sensitive to changes in materials and more fundamentally describe material behavior. If these newer test methods are to be considered for inclusion in material specification, it follows that precision limits for test methods and enforcement limits for construction be established which accurately reflect the expected variation. Work has been underway for four years in the state of Nevada to characterize asphalt mixtures from actual construction projects for indirect tensile strength, resilient modulus, and moisture sensitivity. Materials evaluated for this testing include laboratory prepared and laboratory compacted samples at mix design optimum asphalt content, and loose mixtures sampled during construction and compacted in the laboratory. The data bases were separated first by year (1985 to 1987), then by asphalt grade and asphalt source. Statistical comparisons were used to determine if test results were similar from year to year and between laboratory mixed and laboratory compacted, and field mixed and laboratory compacted samples. The data for these studies have revealed trends which challenge the traditional notion of normality in construction material properties. Statistical evaluation of the data indicates that resilient modulus requires a log transformation to achieve a normal distribution of the data. This finding leads to questions about the calculation of retained strengths for typical moisture sensitivity tests. The typical practice for determining retained strengths is to take the ratio of the resilient modulus after conditioning to the resilient modulus before conditioning. This paper explores the impact on test results and conclusions when the arithmetic ratio is compared to the ratio of the log transformed data. The need for a transformation function for indirect tensile strength is also explored. While data indicates a transformation function could be of benefit, the narrow range of the original data precludes any transformation. This paper includes a discussion of various transformation functions attempted.
Samenvatting