A statistically robust method was developed using the Weibull distribution to identify and eliminate outliers from the failure stress determinations. The method is applicable to any failure stress data set that follows the Weibull distribution; however, in this application, it was developed for the AASHTO standard test method for conducting the direct tension test (DTT). A large number of stress-at-failure measurements with the DTT were made in the course of instructing users of this device. These data, all for the same asphalt, provided the means for studying the nature of the distribution of the breaking strength of these asphalt specimens. The training database contains more than 900 data points. The current AASHTO practice of eliminating the lowest two stress values was found to be reasonable. However, it is an arbitrary method that may lead to problems in the future. On the basis of the results of this study, the procedure is recommended for use and implementation in the next AASHTO version of the DTT standard.
Abstract