Novel FEM-Based Wavelet Bases and Their Contextualized Applications to Bearing Fault Diagnosis

Feature extraction herein refers to using an appropriate wavelet basis to filter vibration signals with the aim to reveal fault transient characteristics, which underlies bearing fault diagnosis. Wavelet transform has developed into a well-established signal processing approach with wide applications in bearing fault diagnosis. Nevertheless, a suitable wavelet basis is essential for wavelet transform to perform its best. So far, numerous wavelet bases are available for bearing diagnosis, most of which, however, have a waveform analogous to that of impulse responses of a single-degree-of-freedom system. In fact, bearings are of multi-degree-of-freedom and not totally rigid. Furthermore, a specific wavelet basis is definitely unable to accommodate all bearing vibrations, given that fault characteristics vary with bearings' operating conditions and fault types. As such, a simulated wavelet-driven personalized scheme is proposed to improve bearing fault diagnosis for contextualized engineering practical applications. For a specific bearing of interest, personalized finite element models (FEM) with various faults are constructed and corresponding fault-induced responses are then obtained. Afterward, FEM-based wavelet bases are formulated and specified by its discrete values from such responses. Taking NU306 bearing with inner or outer defect for example, FEM-based wavelet basis is applied to the corresponding experimental signals by means of wavelet filtering. The comparisons with adaptive Morlet and impulse wavelet demonstrate that the personalized FEM-based wavelet basis match very well with the fault-induced transients present in experimental bearing vibrations and thus have a promising superiority and expandability.