Cacek, Twain2021-09-242021-09-242021-07https://hdl.handle.net/11299/224522University of Minnesota M.S. thesis. July 2021. Major: Civil Engineering. Advisor: Randal Barnes. 1 computer file (PDF); xiv, 119 pages.MnDOT, as part of a multistate pooled fund project, is developing procedures that use ground penetrating radar for QA/QC in the placement of bituminous asphalt pavement. In July 2020, over 400,000 measurements were collected (using the GSSI PaveScan RDM 2.0) from a 5.5-mile stretch of US Highway 2 near Bena, Minnesota. In this thesis, we analyzed that data to develop conceptual models and identify statistical measures, methods, and tests to characterize the results. Graphical representations of our results are presented that will help guide the next phase of field testing with the ultimate goal of developing a national standard. Spatial correlation was characterized using the experimental semivariogram. This analysis revealed that the correlation length of the dielectric (and therefore air voids) ranges from 3[ft] to 45[ft] near the centerline joint, and is less than or equal to 13[ft] in the middle of the lane. These relatively small correlation lengths demonstrate a lack of significant spatial correlation in asphalt density measurements. Lateral variations in dielectric were revealed by partitioning measurements in the road into 1[ft] wide segments. Summary statistics were then computed using the measurements from each segment. The mean of the dielectric values from each segment varied throughout the road. This variation of the mean led to the development of a conjecture that splits the road into three unique zones: the joint zone (i.e., the centerline joint), the transition zone (i.e., the transition between the joint and the mat), and the mat zone (i.e., the middle of the road). This development is important, because current practices in transportation engineering only acknowledge the presence of two zones in the road (joint and mat). The possibility of sensor bias was examined through the use of histograms, probability plots, and statistical tests. Sensor bias was characterized by the prevalence of questionable measurements and dissimilar distributions for each sensor. In this case, the questionable measurements were identified as such because they were either not physically possible, or unlikely to represent an accurate road measurement. The dissimilar distributions demonstrate that, despite sampling the same population, each sensor makes significantly different measurements than the other sensors. Because it is likely present, sensor bias during data collection must be identified and addressed in real time. Recommendations for future data collection and analysis are also provided. These recommendations primarily pertain to sensor bias, data coverage, collection goals, and future experimentation. These recommendations allow MnDOT to move forward in the project knowing that testing procedures are statistically justified.enAsphalt RoadsGround Penetrating RadarMnDOTNondestructive TestingSemivariogramThesis or Dissertation