In-situ recognition of moisture damage in bridge deck asphalt pavement with time-frequency features of GPR signal

Abstract

A complete solution, including an effective non-destructive evaluation (NDE) method and an automatic recognition model, was provided for the rapid diagnosis of moisture damage in the asphalt pavement by using ground-penetrating radar (GPR) signals. A ground-coupled 2.3 GHz antenna was used to conduct a GPR survey on an asphalt pavement of a bridge deck, where the moisture damage areas were detected and visually recognized in processed GPR B-scan images and further validated in subsequent pavement coring. Field GPR traces of the asphalt layer were read and classified to build a dataset which included 8215 moisture damage and 8215 normal pavement traces. A 28-element time-frequency feature vector was extracted and further reduced to an 11-element sensitive feature vector via the linear discriminant analysis (LDA) method. Principal component analysis (PCA) was adopted to decompose the feature vector into the PCs (principal components), which was used to train a BP-ANN model. The result indicates the high accuracy of the ANN model with sensitive feature vectors, i.e., 95.3% for normal and 92.4% for moisture classification. Finally, the ANN model was used to evaluate the GPR survey data, and its result is consistent with the GPR B-scan feature. These findings suggest that the ground-coupled GPR system with 2.3 GHz antenna and the recognition model will enable an innovative quality evaluation system for asphalt pavement.