A Lightweight Fault Diagnosis Framework for Hydro-Turbine Main Shaft Bearing Under Noise Interference

As a critical component of hydro-generating units, the main shaft bearing of a hydro-turbine is essential for ensuring the safety and stability of the unit. However, in industrial environments, operational noise often interferes with fault diagnosis accuracy for main shaft bearings. Thus, this paper proposes a novel noise-robust hydro-turbine fault diagnosis framework. This framework integrates and enhances the local feature extraction capabilities of convolutional neural networks with the global feature extraction capabilities of Transformers. First, Noise-adaptive Random Convolution is employed to randomly perturb the signal, enabling the extraction of noise-robust sample features while preserving the periodic characteristics. Moreover, it introduces only a minimal number of parameters. Second, a parameter-free Light Global Attention mechanism is proposed, which distinguishes key features from noise interference by minimizing the energy differences among similar features. Comparative experiments demonstrate that the lightweight method proposed in this paper exhibits superior diagnostic performance and noise robustness.