As ground-penetrating radar (GPR) is increasingly used for assessing the condition of bridge decks, quantifying and controlling the quality of GPR measures becomes an important challenge. A methodology developed to assess the accuracy of deck condition measures is presented, and its use in a case study involving real data is demonstrated. The latter are generated during GPR applications on a large bridge deck and are processed with a commercial image-processing algorithm. The measures extracted from the processed GPR data are the rebar reflection amplitude and the dielectric constant of the deck material. The accuracy of the GPR assessments is evaluated by comparing core data (ground truth) with the GPR measures. The methodology uses appropriate statistical characteristic curves for quality control. It is based on a use of data to plot the probabilities of true detection versus false detection. Image interpretation requires using a threshold value (typically established from experience) selected to optimize true and false detection rates. The results of the case study indicate that rebar reflection data detect defects of the bridge decks at a 75% true detection rate with a 15% false detection rate. The dielectric data generated during field testing appear not to adequately represent the condition of the bridge deck because of the presence of latex-modified concrete overlay. The details of this finding and important conclusions are presented and discussed.
Abstract