Three field test locations (Texas, Virginia, and Nevada) were used to produce a data base of more than 900 nuclear density readings to investigate the precision of the American Society for Testing and Materials test method 2950. A combination of private, state, and county laboratories throughout the three states, as well as three gauge manufacturers, provided a total of 31 different gauges. Each field location consisted of 10 test sites and at least two different hot mix asphalt pavement conditions. Participating laboratories at each location tested the same test sites using 15-sec, 1-min, and 4-min readings. Test sites were cored after the nuclear density readings had been taken. Statistical analysis of the data showed that a 15-sec reading generated a similar density reading to either the 1-min or the 4-min readings. A two-way analysis of variance showed that all gauges and test sites were significantly different. Further statistical analysis showed variances generated by each test location to be dependent on each specific set of test conditions. Regression equations were developed for each gauge for each test location; nuclear density readings were correlated to densities determined from the bulk specific gravities of the corresponding cores. When considered as a group, gauges fail to generate an accurate regression equation. When considered individually, however, the gauges are capable of producing an r-squared of 0.8 Or greater. Regression equations also appear to be dependent on test conditions. Correlations between r-squared and standard counts, date of last calibration, and average differences between cores and gauges showed no apparent trends.
Abstract