Current-based Gearbox Fault Diagnosis Based on Sparse Filtering Feature Fusion

Vibration signal-based analysis is the main fault diagnosis method for wind turbine gearboxes. Compared with the vibration signal, the current signal has the advantages of non-invasive and low monitoring cost. Therefore, a fault diagnosis approach for wind turbine gearbox based on the generator current signal is proposed. The current signal usually presents a large interference of fundamental frequency component and a low signal-to-noise ratio, leading to difficulties in feature extraction. In order to address these issues, an unsupervised feature learning and fusion method using current signals based on a two-layer sparse filtering network is proposed. Firstly, a local feature learning network based on sparse filtering is designed to learn different fault features from the original current signal and the envelope signal respectively. Then, the sparse features of the original signal and the envelope signal learned in the sparse filtering network are fused with the aim to enrich the fault feature space. Finally, the fused features are fed into the support vector machine for training to realize the intelligent recognition and diagnosis of different fault types. The proposed method is validated through the gearbox faults experiments on a wind turbine gearbox test rig. Experimental results show that the proposed method can automatically extract the useful features reflecting gear faults from the current signals. Compared with the traditional feature extraction method, it achieves higher diagnostic accuracy and efficiency. © 2020, Power System Technology Press. All right reserved.