Abstract

Bridge infrastructure requires periodical inspection, repair-maintenance, or complete replacement of bridges for long-term safety and sustainability. The visual inspection method has been predominantly used by transportation agencies for bridge condition assessment whereby qualified bridge inspectors visually observe the surface defects and based on their knowledge, scale the bridge components according to their condition. However, this method does not provide accurate information about the bridge condition due to its inability to detect internal defects. While various non-destructive techniques are being increasingly adopted for wholesome condition assessment, Ground Penetrating Radar (GPR) is recommended for bridge evaluation due to its distinct advantages of identifying major subsurface defects rapidly. The interpretation of data obtained from GPR profiles is a major issue for bridge inspectors due to inconsistent correlation with the actual bridge condition. Federal agencies commonly employ an amplitude-based approach that involves picking amplitude values at specific locations in a GPR profile which are indicative of concrete corrosion levels. However, this approach may not always yield reliable results as it disregards the majority of the information present within a GPR profile. A novel approach based on image-based analysis involves an experienced analyst reviewing the GPR profiles and marking attenuated areas across them while considering structural and surface anomalies, and other several parameters, which are typically ignored in amplitude-based analysis. Thus, this holistic method generates less noisy condition maps and has demonstrated good correspondence with the actual ground condition. However, this approach is highly subjective, dependent on the level of knowledge of the analyst, and time-consuming. To overcome these shortcomings, the image-based approach of analyzing GPR data has been automated based on scientific analysis and image processing tools by evaluating the GPR profiles as images since most of the information contained in these profiles can be better analyzed visually. Two successful models have been developed which generate reliable condition maps. The first comprehensive model is based on Viola-Jones Algorithm, and it identifies hyperbolic regions automatically with a trained classifier to develop condition maps using entropy values of detected regions and K-means clustering. The second model is relatively a more robust and wholesome approach for analyzing GPR data as it considers user-assisted information, identifies complex hyperbolic sig-natures based on downward openings and local maximums of the distance transform image, detects anomalies, and assigns an optimal cluster value to generate condition maps. As part of the validation, both these methods were implemented on real case studies that have shown good resemblance with the existing amplitude-based approach, image-based analysis, and more importantly, correlated with the destructive coring samples. Finally, lab tests were conducted in a controlled environment to evaluate the measured factor, entropy, used for detected regions and compare it with the amplitude values for subsurface objects and the results indicated that entropy yielded comparatively better results for all cases (rebars, air gaps, and water gaps). Consequently, condition maps generated based on developed models can be efficiently used by bridge inspectors in making informed decisions regarding the repair and rehabilitation of reinforced concrete bridges.