Multi-view feature encoding model with wavelet transform embedded for intelligent machine fault diagnosis

Intelligent machine fault diagnosis technology is irreplaceable in modern industrial production. Tremendous development has been made in recent years in researching deep learning (DL) based diagnosis methods. But most current DL-based methods still face the challenge, which does not consider the sufficient information between different views of signal and incorporate specific knowledge. To address this issue, this paper proposed a multi-view feature encoding model (MVFEM) by embedding the wavelet transform knowledge to realize the fault diagnosis of rotary machine. To this end, the proposed MVFEM 1) utilizes multi-branch wavelet convolution via using different wavelet basis functions to capture the multi-view features between different wavelet character views, 2) feeds the multi-view feature into a temporal-spatial block to mine the temporal-spatial related information and 3) fuse the multi-view feature as one feature for diagnosis. With the capability to mine the complex relation from different views of individual vibration signal, the proposed MVFEM can leverage the information from multi-view features to monitor the condition of the rotary machine and detect the faults. The performance of the MVFEM in rotary machine diagnosis is validated on the experiment and compared with other related methods. Experimental results illustrate that the proposed MVFEM can learn useful multi-view feature to realize accurate fault diagnosis.