Falling Weight Deflectometer (FWD) data and the corresponding pavement stiffness moduli and deflection basin areas vary as functions of both time and space. This paper focuses primarily on extracting information from the spatial variability of FWD deflection data. Spatial variability occurs both horizontally and vertically within a pavement system; it is inherent in the system due to the heterogeneity of the material composing the subgrade. and is further influenced by the construction process and the resulting variations in density, moisture content, and thickness of the subbase, base, and surface layers.
To assess spatial variability of just the subgrade and then the overall pavement structure, FWD tests were conducted on 71 test points on top of each of the layers composing a pavement system. and a statistical analysis was conducted on the data. Test pavements included two mainline (five-year design) and two low-volume test cells at the Minnesota Road Research Project (MN/ROAD).
By comparing differences between pairs of measured deflections at increasing test point separation distances. one can incorporate distance weighting techniques into a statistical form frequently used in the fields of geology and mining. The geostatistical semi-variogram can be applied to pavements and used to model the degree of correlation between data at any two test points. As the distance between test points increases, corresponding data become decreasingly dependent upon each other until, at some appreciable distance. they are independent of each other. From the semi-variogram. one can readily determine the separation distance at which values are independent of each other. Conventional statistical analyses are also used to supplement geostatistical techniques.
Valuable and cost-saving information can be acquired by analyzing baseline FWD data with this technique. The efficiency of future FWD testing can be maximized, and optimum FWD test point spacing can be determined for pavement evaluation and overlay design. Furthermore, the geostatistical techniques discussed are applicable to any problem involving the distribution of a variable in one. two, or three dimensions.
