In this study, we present an approach of using Artificial Neural Networks (ANNs) to predict the locations of cracks in structural elements. The model utilizes input features derived from the Relative Frequency Shifts (RFSs) of six out-of-plane bending vibration modes. The proposed approach uses a stacking technique that integrates multiple Multi-Head Recurrent Convolutional Neural Networks (MH-RCNNs) as base models and a meta-model that aggregates their predictions. This method effectively combines the strengths of neural networks and ensemble learning. The dataset used for training includes computed RFSs from various damaged locations, providing a comprehensive foundation for the task. The base models are trained on a dataset representing the entire test beam. Several tested meta-models then aggregate the predictions from these base models to make a final prediction regarding the location of damage. This stacked architecture enhances performance, providing a reliable method for accurately predicting damage. The reliability of the proposed AI-supported method was effectively demonstrated using RFSs derived from numerical simulations. This robust approach highlights the method’s effectiveness in real-world applications, ensuring its credibility and practicality.