A Novel Physics-Based Polar Diagram for the Diagnosis of Compound Faults with Similar Characteristics

Compound fault diagnosis of rotating machinery, including bearing, rotor and gear, has always been attracting increasing attention from researchers. Among them, the classification and isolation of compound fault with similar characteristics remains a challenging issue. To this end, a physics-based polar diagram with convolutional neural network (CNN) model is proposed to distinguish the similar fault characteristics by fusing the self-sensing motor drive signals. Firstly, the misalignment and bearing fault model are constructed, and their impacts on q-axis current in motor drive are derived. Secondly, the empirical mode decomposition (EMD) is used to extract the feature-related components from the q-axis current signal. Subsequently, an improved physics-based polar diagram (PPD) is constructed using the feature-related q-axis current components and modified motor phase signal with physical knowledge and data features. Finally, comparative experiments with a baseline CNN shows that the PPD proposed in this paper is superior to traditional image representation methods in precision and transfer learning ability.