Learning the Appropriate Local Averaging Length for Traffic Speed Deflectometer Deflection Velocity Measurements: Application to the Analysis of Rigid Pavement Joint Responses

Abstract

The dilemma of selecting an appropriate averaging length for traffic speed deflectometer (TSD) measurements has not yet been satisfactorily resolved. Average too much, and important structural features are lost. Average too little, and the data are plagued with noise that can be confused with structural features. In addition to the challenge of selecting an appropriate averaging length, structural feature variations can occur at different length scales. On relatively uniform sections, these occur at scales that are in the tens, hundreds, or even thousands of meters; around the joints of a jointed concrete pavement, variations occur at the centimeter scale. In this paper, we present a method that adaptively averages 5-cm resolution TSD deflection velocity measurements and efficiently removes the noise in the measurements while preserving the structural features of interest. The approach is based on classifying changes in the measured pavement response as either changes in the structural response or changes caused by measurement noise. The classifying threshold is determined by minimizing the (weighted) classification error. We present a simulated example to illustrate the proposed approach and then use it to analyze TSD data collected at the MnRoad low volume road testing facility. The results show that the proposed approach preserves interesting structural response features at the joints of a jointed concrete pavement while at the same time smoothing the measurements and removing the noise at locations where the structural response is relatively uniform. The results are validated using a mechanistic model of a beam on a Pasternak foundation.