A Noise-Robust Fault Diagnosis Algorithm for Rotating Machinery with Limited Sample Size

To address the issues of noise interference and the scarcity of fault samples in the complex environments of industrial sites, this study proposes a small-sample bearing fault diagnosis framework robust to noise disturbances. A noise processing module through Empirical Mode Decomposition is designed to effectively filter out noise signals from the collected data. Subsequently, a data transformation module converts time-domain signals into frequency-domain signals for further analysis. Next, a diffusion model is employed to enhance samples, thereby mitigating the limited availability of fault instances. Finally, a fault diagnosis module leveraging Convolutional Neural Networks is constructed to achieve high-precision fault classification. Simulation experiments using the dataset from Case Western Reserve University demonstrate that, compared to existing multi-classification algorithms, the proposed method significantly improves noise robustness and fault classification accuracy under small-sample conditions.