Attention Mechanism Feature Fusion for Bearing Fault Diagnosis in Noise Environment

Rotating machinery is a crucial component and plays a vital role in ensuring system safety and stability. Fault diagnosis in such systems can be challenging due to environmental noise. Previous research has effectively improved fault diagnosis performance in noisy environments by enhancing the quality of feature representations. This paper introduces an innovative approach for fault diagnosis, named Attention Mechanism Feature Fusion (AMFF), employing an attention mechanism to construct two distinct features. These features, namely Mel-frequency Cepstral Coefficients (MFCCs) and Fast Fourier Transform (FFT), respectively capture time-evolving spectral characteristics and frequency-domain data. The channel-spatial attention mechanism highlights salient features, minimizes redundant information, and thereby mitigates the noise impact. The method can effectively extract high-quality features and maintaining precise classification performance in noisy environments, as demonstrated through experimental validation. This study has far-reaching implications for improving diagnostic capabilities in noisy environments and providing potential for numerous industrial applications.