Detection and Assessment of Cracks in Concrete Structures Using Machine Learning Techniques: A Review

https://doi.org/10.46610/JoCCS.2026.v011i02.001

Abstract

Cracks represent a common manifestation of concrete deterioration. The concrete construction exhibits fissures at the microscopic level. A consistent change in the structure’s size results in its failure. Crack screening methodologies encompass conventional, optical, and asymmetrical screening approaches. The conventional method evaluates the divisions through a rudimentary graphic that depicts the different states of the variances. The visual method depends on human beings to identify fractures. It is an amalgamation of human perceptual abilities and proficiency. Moreover, manual inspection is primarily employed in developing countries for the detection of fractures. It utilises scanning and tactile devices to identify and delineate fractures. Nonetheless, these methodologies possess specific constraints, like the necessity for a trained practitioner, the degree of expertise, the machinist’s understanding, and the resolution of the images. Researchers conducted multiple investigations to accurately detect fissures in the material’s framework. They have advanced the methodologies by employing image processing techniques, including edge recognition, segmentation, and categorisation. Crack detection procedures are classified into geographical, computational learning (ML), and deep learning (DL) image processing techniques (IPAs). The literature indicates that the most often employed image processing algorithms (IPAs) are the Sobel filter, Canny edge detector, Roberts’ operator, Prewitt operator, and Otsu’s threshold-based method. The efficacy of these methods is contingent upon the texture, noise, and quality of the photo. Furthermore, these techniques depend on the choice of cortical masks and sensitivity parameters. The effectiveness of machine learning-based crack detection methods relies on deciding on a set of handmade features and a precise division of the region of interest. Furthermore, the machine learning-based methodologies necessitate substantial human involvement. Moreover, DL-based approaches require accurate annotation for efficient crack detection. The research indicates that immediate crack evaluation requires the creation of a GUI for analysing the condition of ceramic structures.