Multivariate hierarchical multiscale fluctuation dispersion entropy: Applications to fault diagnosis of rotating machinery

With a view to completing the fault diagnosis of rotating machinery efficiently and accurately, this paper presents a novel fault diagnosis model that combines multivariate hierarchical multiscale fluctuation dispersion entropy (MHMFDE), multi-cluster feature selection (MCFS) and gray wolf optimization-based kernel extreme learning machine(GWO-KELM). Firstly, MHMFDE is presented to capture the high-dimensional fault features hidden in the attained multichannel vibration signals. Integrating the multiscale entropy-based method and the hierarchical entropy-based method that are currently popular in the domain of fault identification, MHMFDE can simultaneously extract affluent fault features from multivariate vibration signals in-depth as well as overcoming the problem of information loss in the existing single-channel data analysis methods. Afterward, MCFS is used to pick sensitive features from the attained raw fault features to form the sensitive feature vectors, thereby reducing the impact of redundant features. Finally, GWO-KELM is adopted to quantitatively analyze the diagnostic effect. Three examples reveal that the presented approach enjoys excellent performance in the fault diagnosis domain. Especially for the identification of compound faults of rotating machinery, the performance of the presented method is significantly superior to that of existing methods. (C) 2021 Elsevier Ltd. All rights reserved.