A Robotic-Based Framework for Quantifying Surface Cracks of Concrete Shear Walls

Abstract

The inherently subjective and manual nature of current structural assessment procedures undermines the reliability of such practices. As a result, researchers are using artificial intelligence and robots to encourage autonomy and create methods that are more objective and less biased. Robots are equipped with cameras and sensors to collect information about the structure, such as images of the crack patterns of concrete structures. In this paper, a robotic framework is proposed to utilize the collected images of reinforced concrete shear walls (RCSWs) with crack patterns, and then those images are used to quantify the level of damage. Crack pattern images are from the load steps of a quasi-static test on an RCSW. To quantify the extent of damage, crack patterns are first converted to a mathematical representation, a graph. Next, a machine learning algorithm is trained to predict the energy dissipated during each load cycle based on the graphs. Results reveal that the presented graph-based damage quantification method is able to quantify the level of structural damage. High R2 scores of the machine learning regression (above 0.98) attest to the success of the proposed robotic-based framework.