Detection of cracking mechanism transition on reinforced concrete shear walls using graph theory

Abstract

Concrete crack quantification is a crucial step toward the assessment of concrete structures. Although many computer vision algorithms have been developed to detect cracks and measure their properties, such as width and length, interpretation of the detected cracks in terms of structural behavior remains a challenge. Specifically, identifying the onset of changes in the behavior mechanism (e.g., shear or flexural) of the structure is of great interest. This is particularly important in concrete shear walls subject to cyclic loading, in which cracks may close and thus cause crack width measurements to be unreliable. In such structures, individual and disjointed cracks gradually form mosaic patterns. The transition from one state to the other results in a sudden change in the cracking patterns. This study builds upon the previous work of the authors and uses graph theory to represent concrete crack patterns. The main idea is to utilize graph features and their changes to track changes in the crack patterns. To validate the proposed method, surface crack images of 15 large-scale reinforced concrete shear walls under cyclic loads are used. For each wall, the images include crack patterns at different load levels. Using the proposed methodology, the images of the crack images are converted to their representative graph. Afterward, two specific graph features are extracted: 1) the average degree of network (k_avg) and 2) the weighted average degree of network (kw_avg). The ratio of k_avg/kw_avg versus the drift of the walls at each load cycle is used to detect the change in the cracking mechanism. Results show that the minimum value of the ratio corresponds to the change in the cracking mechanism. The robustness of the proposed metric indicates that it can be used for training machine learning models to develop systems that can automatically signal the onset of a change in the cracking mechanism.