Evaluation of Spalling in Bridges Using Machine Vision Method

The growing number of bridges and their deteriorated conditions on one hand and the budget squeeze for their repair and rehabilitation on the other call for automated detection of defects and smart methods for condition rating of these bridges. This paper presents a newly-developed standalone computer application for automated detection and evaluation of spalling severities in reinforced concrete bridges. The application is coded in C#.net and makes use of an early developed model for detection of surface defects. The method is applied in two tiers, in the first tier, a single-objective particle swarm optimization model is developed for detection of spalling based on Tsallis entropy function. The second tier is devised for evaluation of spalling severities. It generates a comprehensive representation of the bridge deck image using Daubechies discrete wavelet transform feature description algorithm. The second tier also encompasses a hybrid artificial neural network-particle swarm optimization model for accurate prediction of spalling area; circumventing the drawbacks of the gradient descent algorithm. The developed method was tested using 60 images from three bridge decks in Montreal and Laval in Quebec, Canada. Results indicate significant superiority in area prediction accuracies; achieving mean absolute percentage error, mean absolute error and relative absolute error of 6.12%, 56.407 and 0.393, respectively. The developed method is expected to assist transportation agencies in performing more accurate condition assessment of concrete bridge decks and accordingly assist them in developing optimum maintenance plans.