Abstract

Abstract
Image-based Condition Assessment for Concrete Bridge Inspection

Ram Sebak Adhikari, Ph.D.
Concordia University, 2014

The following approaches are usually taken for the condition assessment and performance evaluation of civil infrastructure: visual inspection, structural response measurement due to loads, and sensing based inspection of bridge structures. This thesis concentrates on the last alternative using remote sensing for condition assessment of concrete bridge structures. Focusing on defect quantification problems for condition assessment of bridge structures, remote sensing techniques based on digital images provides superior result over conventional visual inspection-based methods. The aim of this thesis is to develop digital image-based condition assessment tools and techniques, which can be integrated with existing bridge management systems (BMSs) in order to enhance the reliability of current inspection practices.

The methodology of this research divides the entire task of bridge inspection into two modules. The first module develops quantification models based on the extent and severity of defects, and the second module develops a change detection model defined as change in element condition state over times. For defect quantification, three fundamental concrete defects such as cracks, spalling, and scaling have been considered. To illustrate the proposed methodology, digital images are acquired from laboratory experiments during the testing of reinforced concrete beams in flexure, and from field visits of bridges in Montreal, Quebec using portable digital cameras. This research contributes in the development of crack quantification model based on the corresponding crack skeleton which takes consideration of crack tortuosity for retrieving of crack properties. The output of the crack quantification model is validated by capturing the crack properties using a crack scale. In addition, an automated model for estimating the condition rating and related computational algorithms for bridge inspection are developed using the guidelines of the Ontario Structure Inspection Manual. The developed algorithms for mapping of condition ratings are based on the supervised training of back propagation neural networks. Recognizing the importance of 3D visualization, which can mimic the on-site visual inspection, 3D visualization model is developed using ordinary digital images by manually projecting images on the 3D model of the bridge being inspected.

The second module proposes a novel approach for periodic detection of defects in concrete bridges based on a set of dimensionless metrics pertinent to spectral and fractal analyses of the captured images. The fractal analysis of digital images is described by fractal dimension (FD) using Box Counting algorithms. Similarly, the method of spectral analysis requires digital images to be translated from spatial domain to Fourier domain, and then finds one dimensional signatures to quantify change detection. The developed algorithm for change detection demonstrates superior results and eliminates the limitations of traditional approach of change detection based on image subtraction. The developed image-based models can either be applied as standalone condition assessment and rating applications or integrated with existing systems such as PONTIS ( a Bridge Management System in USA) in order to enhance the reliability of visual inspection.