Ensuring the structural stability of dams is critical to prevent failures and guarantee their safe long-term operation. This study investigates the use of Siamese Neural Networks (SNNs) with 1D convolutional architectures for anomaly detection in dam structural monitoring data. Unlike conventional classification models, the SNN is designed to learn a similarity function, allowing it to generalize beyond known damage patterns and accurately detect previously unobserved anomalies. The model is trained using a one-class learning approach, considering only normal conditions and a subset of anomalies during the training phase. The evaluation covers both previously seen anomalies and unobserved scenarios, reflecting real-world conditions where new structural failures may emerge. The results show that the SNN maintains high accuracy across all test cases, achieving perfect classification (100.0%) under normal conditions and near-optimal performance in the most challenging scenarios. In contrast, traditional methods exhibit considerable variability, especially when attempting to classify anomalies that are not represented in the training data, evidencing their limitations in addressing novel scenarios. By leveraging 1D convolutional architectures, SNN effectively captures spatiotemporal relationships in sensor data, strengthening its ability to differentiate between normal and faulty conditions. These findings highlight the potential of deep learning to improve structural health monitoring (SHM) strategies in dams, providing a robust framework for real-time anomaly detection and early warning system deployment in critical infrastructures.