Multiview enhanced fault diagnosis for wind turbine gearbox bearings with fusion of vibration and current signals

Currently, most fault diagnosis methods for wind turbine gearboxes rely on certain unimodal signal, such as vibration or current, which cannot enable reliable and satisfactory performance due to its limited presentation ability. To this end, this paper proposes a new multiview enhanced fault diagnosis framework to learn the correlated and complementary features across current and vibration signals, which are regarded as two different but related views. Multiple statistic features at different wavelet packet decomposition levels are first extracted from raw vibration and current signals, respectively. Then, an unsupervised multiview learning method based on canonical correlation analysis (CCA) is developed to learn maximum correlations between vibration and current features. Finally, the learned enhanced features are used to identify different health conditions. Experimental results show that our proposed method can learn enhanced fault-related features and achieve superior fault diagnosis performance, especially on compound faults, compared with unimodal signal-based methods.