Deep learning for crack detection in concrete structures: A systematic literature review
Year: 2026
Type: review
Study design: Systematic Review
Abstract: Early crack identification is essential for improving maintenance planning and reducing the risk of premature deterioration in structures. This systematic review examines recent applications of deep learning to these tasks and compares the architectures used, data sources, preprocessing strategies, reported performance, and limitations on field deployment. Scopus and ScienceDirect were searched for the 2022–2026 period. The initial search retrieved 7,132 records from both databases: 572 from Scopus and 6,560 from ScienceDirect. After filters for subject area, document type, and open-access availability were applied, 1,303 publications remained; relevance screening produced a final set of 71 studies, including 25 from Scopus and 46 from ScienceDirect. Metadata from the 108 filtered Scopus documents were analyzed using the open-source software VOSviewer to identify relationships among keywords. The results show a shift from convolutional models focused on classification toward semantic and instance segmentation architectures, vision transformers, and adaptations of foundation models such as SAM. Increasing integration of RGB images, ultrasound, radar, acoustic sensors, UAVs, IoT systems, and three-dimensional reconstruction with digital twins was also observed. Preprocessing, super-resolution, and sensor fusion directly affect diagnostic quality. Nevertheless, two challenges persist: performance declines when models are transferred to conditions different from those used for training, and their predictions remain insufficiently explained. The review concludes that the widespread adoption of these tools still requires external validation, benchmark data sets, comparable protocols, and joint evaluation of accuracy, robustness, and computational cost. Explainable artificial intelligence and integration with digital twins are priority directions for turning automated detection into reliable support for structural management.
Source: Multidisciplinary Reviews
Authors: Quispe Santillán, Telmo José, Pérez, Sócrates Pedro, Garcia Chumacero, Juan Martin, Ruiz Pico, Angel Antonio, Valdiviezo Sir, Victor Manuel, Carranza Guevara, Rosas, Merino Cava, Luis Gerardo, Medrano Lizarzaburu , Eithel Yván, Sevilla Gamarra, Marco Antonio, Espinoza Rodriguez, Olenka Ana Catherine, Villena Zapata, Luigi Italo
Institution: National University Toribio Rodríguez de Mendoza
Language: English
Subfield: Civil and Structural Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7220156017