Weighted Multi-view Zero-shot Learning Prototype Shift Model in Fault Diagnosis

Different fault severities in the mechanical fault diagnosis field contribute to different fault classes' distribution and prototype position. It brings a challenge to mechanical fault diagnosis. This paper proposes a weighted multi-view hierarchical, zero-shot learning model to address the problem. The multi-view inputs are proposed to locate the positions of fault prototypes accurately. To precisely utilize multi-view input information, an unsupervised weighting strategy is proposed to balance positive and negative views. Hierarchical classification is utilized to gradually change this classification boundary or prototype position of the model. The effectiveness of the proposed method is verified through comparative experiments, and the results indicate that the model can classify mechanical fault across different fault severity problems. Compared with other classical zero-shot learning models, the results show that the proposed model performs better in mechanical fault diagnosis than the existing methods.