Fault diagnosis method of planetary gearbox under heavy noise based on time- and frequency-domain feature enhancement fusion

Abstract

Planetary gearboxes, serving as critical components in wind turbines, are often subjected to complex and noisy operating environments that pose considerable challenges for reliable fault diagnosis. Traditional time-domain methods are highly sensitive to noise, whereas frequency-domain approaches may fail to capture essential time-localized fault information. To overcome the limitations of single-domain analysis and improve diagnostic reliability in real-world engineering applications, this paper proposes a novel fault diagnosis method based on time–frequency domain feature fusion with mutual information loss. Firstly, the constructed residual connection structure based on cross-domain feature enhancement extracts features from both the time and frequency domains by the residual structure for cross-domain fusion to maintain the compactness of the feature space. Secondly, the designed adaptive feature fusion mechanism is employed to adaptively fuse time and frequency domain features, while the proposed feature correlation enhancement strategy is introduced to strengthen the correlation between time and frequency domain features. Finally, the proposed method is validated on fault datasets collected from two different types of planetary gearboxes. Experimental results demonstrate its superior diagnostic accuracy and robustness under strong noise conditions, highlighting its practical potential for intelligent condition monitoring in wind energy systems.