Hierarchical discriminating sparse coding for weak fault feature extraction of rolling bearings

The effective extraction of weak fault features is crucial in condition monitoring and fault diagnosis of rolling bearings. Sparse coding, as a promising tool for signal denoising and feature extraction, has attracted a lot of attention in recent years. However, many challenges still exist when sparse coding is applied to the bearings detection under harsh working conditions. Specifically, the predefined dictionary-based sparse coding (PDSC) usually needs prior knowledge about the target signal, while the learning dictionary-based sparse coding (LDSC) is susceptible to interfering components produced by other rotating parts, thus bringing difficulties for early fault identification. To overcome these disadvantages, a hierarchical discriminating sparse coding (HDSC) method is presented in this paper, which could process the raw signals directly and utilize hierarchical concept to isolate interferences. In addition, a novel index termed envelope harmonic-to-noise ratio (EHNR) is introduced to give the instruction on reasonably choosing the parameters in the process of HDSC. The advantages of HDSC over traditional approaches are validated on the simulated signals and real vibration data from locomotive bearing. The results demonstrate that the proposed method can successfully extract the weak fault feature even in the presence of strong noise and ambient interferences.