Vibration-based deep learning framework for predicting crack evolution in reinforced concrete slabs

The ability to identify structural damage at an early stage is essential to ensure the structural integrity, maintenance, and performance. This paper analyzes a reinforced concrete slab using a finite element model, in which the material nonlinearity is considered. Through incremental loading combined with transient analysis and stiffness updating, the evolution of cracks, element states, and modal characteristics at each load level are extracted and used as input for the proposed damage assessment approach. Then, a two-step framework is developed: the first step focuses on damage localization using the modal strain energy-based technique, while the second step quantifies damage severity using artificial neural networks trained on modal features. The damage assessment process is carried out continuously during the development of cracks under loading, and a confusion matrix is used to measure the performance of the proposed method. In addition, a shape characteristic of the modal strain energy-based index is newly introduced to mitigate noise effects when identifying the undamaged state. The results demonstrate that the use of only a few bending modes is sufficient to significantly improve prediction accuracy, and the proposed approach effectively identifies the occurrence, location, and severity of cracking in reinforced concrete slabs under loading conditions.