Multivariate embedding multiscale diversity entropy by multi-channel signals for fault diagnosis of rotating machinery

The multivariate entropy methods enhance fault diagnosis accuracy in rotating machinery by comprehensively analyzing multi-channel signals. However, traditional multivariate strategies have not determined the phase space reconstruction parameters for different channels, which directly damages the complexity quantification results of entropy values and thus impairs the fault identification results. In this study, multivariate embedding multiscale diversity entropy (MEMDE) is innovatively proposed. It constructs accurate phase space structures for signals from different channels, avoiding low correlation signals from being reconstructed into multivariate phase spaces, thereby improving the accuracy of entropy estimation. A comparative analysis and validation of the proposed method were conducted on multi-channel rotating machinery data collected from centrifugal pumps under different fault states. The results demonstrate that MEMDE has the highest accuracy, efficient multi-channel fusion, and strong nonlinear feature extraction and recognition capabilities in fault identification.