부산대학교 도서관

Addressing the problem that the pulse component often submerges in other frequency components during the failure of the wind turbine gearbox, it is difficult to extract effectively the early fault characteristics. This paper proposed autoregressive minimum entropy deconvolution enhanced resonance features for wind turbine gearbox fault diagnosis. The method is combined with 1DCNN to achieve high-accuracy fault diagnosis. Firstly, the resonance-based sparse signal decomposition was utilized to decompose the vibration signals into a high resonance component containing noise and harmonic components and a low resonance component containing fault impulse components. Secondly, the AR-MED was employed to enhance the weak impulse features in the low resonance component, thus further enhancing the periodic impulse components of the gearbox. Finally, a feature enhanced 1DCNN was constructed to fuse the decomposed harmonic and periodic impulse components for feature integration, targeted training, and classification. The outcomes of the experiment indicated the efficacy and supremacy of the suggested approach in extracting information about fault characteristics and enhancing the accuracy of fault diagnosis in wind turbine gearbox systems in contrast to current fault diagnosis models.