Abstract: Fault diagnosis is essential for ensuring the safe and efficient operation of hydroelectric generating units. In recent years, data-driven diagnosis methods have gained wide attention, but traditional approaches often rely on manual feature extraction and selection, which limits their performance. This paper proposes a fault diagnosis method based on signal-image mapping encoding and deep convolutional neural networks (DCNN). One-dimensional vibration signals are first converted into two-dimensional images, which are then input into a DCNN to automatically extract fault features and train the diagnosis model, avoiding the subjectivity of manual feature selection. Experiments on three datasets compare different mapping methods, denoising strategies, and network architectures. Results show that the proposed method outperforms traditional approaches in feature learning, fault identification accuracy, and noise robustness, demonstrating its theoretical and practical value.