MCrack-Dam: the scale-up of a method to assess cracks on concrete dams by image processing. The case study of Itaipu Dam, at the Brazil–Paraguay border

Abstract

Inspection and maintenance of civil infrastructures require structural assessment, usually performed based on the monitoring of critical sections. For concrete structures, the identification and characterization of the crack patterns is an important task to a rigorous evaluation of the structural performance. In the case of concrete dams, the timely detection and correction of structural problems can avoid major accidents. Despite the significance of crack monitoring and the recent innovations using image processing, the inspection of dams is usually simply based on visual inspections. This results in sketching crack patterns and also includes hand-held measurements, using crack width rulers and measuring tape. Thus, the development of automatic methods based on image processing to assess cracks in concrete dams has significant advantages. In this scope, most of the methods were applied in the laboratorial environment, and a gap to scale-up them to onsite assessment is clearly identified. In this paper, a method named MCrack-Dam, resulting from the scale-up of the method MCrack, previously developed and validated in controlled laboratorial conditions, is presented. The method is based on image processing and designed to automatically monitoring cracks in concrete dams. The MCrack-Dam relies on a predefined systematic acquisition of images: pre-processing those images for ortho-rectification; processing of the latter to identify and model cracks; and post-processing procedure to characterize the crack key parameters. The method was applied to a predefined region of the Itaipu Dam, at Brazil–Paraguay border. The results validate the ability of MCrack-Dam for performing a detailed characterization of cracks in concrete dams, not comparable to the traditional methods currently used. In addition, MCrack-Dam successfully works on surfaces with distinct features (‘noise’ for crack detection) such as drippings, sketched drawings, and smooth and rough textures, unlike other image processing methods when applied on ‘noise’ surfaces. Finally, the most relevant conclusions, and guidelines for the optimization of the exhaustive survey of the entire surface of the dam, are presented.