A robust entropy-based sparsity measure for health monitoring of rotating machinery

This study presents a novel negative entropy-based sparsity (EnS) measure for enhancing condition monitoring of rotating machinery. The proposed sparsity measure, derived from a probabilistic entropy framework, offers a superior method for quantifying signal sparsity and tracking degradation in lifetime failure data. Existing sparsity measures such as Shannon entropy, kurtosis, Hoyer measure, and L2/L1 norm fall short of satisfying critical properties of a sparsity measure such as Robin Hood, Bill Gates, cloning, and Rising Tides. The EnS measure proposed in this work, however, fulfills all six essential attributes of a sparsity measure, making it a promising candidate for condition monitoring applications. The proposed measure is particularly effective in analyzing intelligent maintenance systems data, demonstrating improved robustness compared to negative Shannon entropy-based measure. The quantitative evaluation, based on the fluctuation of the sparsity measure and the coefficient of variation under heavy noise conditions, demonstrated that the EnS measure is a promising candidate for effectively capturing the intrinsic dynamics of the degradation process. Overall, the EnS measure advances machine health monitoring techniques, offering a more reliable method for proactive maintenance.