Cross Diversity Entropy-Based Feature Extraction for Fault Diagnosis of Rotor System

The rotor system in the fault state generally shows apparent nonlinear behavior and complex dynamic characteristics. In general, the data collected from multiple sensors (namely, multichannel signals) is required to achieve accurate condition monitoring of the rotor system. Even though the traditional multivariate entropy method can extract fault characteristics of the vibration signals from multiple sensors in the rotor system, some critical fault information might be lost when it is applied to process the multivariate information. In this article, a novel approach called cross diversity entropy (Cross-DE) is proposed to address the issue discussed above. More specifically, when processing the multichannel signals, the developed Cross-DE uses the diversity of orbits in different phase spaces to represent the complexity of the whole system. Moreover, in the process of calculating the diversity of orbits, Cross-DE calculates the cosine similarity between the orbits in the same phase space and different phase spaces, resulting in information interaction between different channels. Furthermore, the developed Cross-DE fuses the information from different channels to solve the problem of fault information loss during feature extraction. Therefore, the developed Cross-DE has the capability of accurately extracting complete fault features from multichannel, which benefits the health management of rotor system condition monitoring significantly. The 2-D displacement signal of the rotor is applied in this article to demonstrate the performance of the Cross-DE. Also, the effectiveness of Cross-DE is verified using both simulated signal and experimental data.