Structural Damage Identification of Hydraulic Structures via Multisource Data Fusion and Deep Learning

Structural health monitoring (SHM) is vital for ensuring the safe operation of hydraulic structures. Traditional single-sensor systems often face limitations such as incomplete data and high false alarm rates. To overcome these issues, this study proposes a deep learning-based damage identification method that integrates multi-source monitoring data, including vibration signals, strain measurements, and acoustic emission data. A multi-channel convolutional neural network (CNN) is constructed to extract features from each data type in parallel, followed by feature fusion and classification. The model is validated using real-world monitoring data collected over 12 months from a large-scale concrete gravity dam. Experimental results show that the proposed method achieves a classification accuracy of 94.267% across five damage levels, outperforming both single-source models and traditional machine learning approaches. Furthermore, sensitivity analysis of hyperparameters and fusion weights reveals the relative contribution of each data type, highlighting the effectiveness of combining global, local, and micro-scale damage indicators. This study demonstrates the potential of multi-source data fusion and deep learning techniques in enhancing the accuracy, robustness, and automation of structural damage detection in hydraulic infrastructure.