Answers are grounded in this paper and cite their sources.
Phase-space reconstruction supplies the geometric foundation of nonlinear vibration analysis, yet its delay parameter is sensitive to the impulsive glitches and local extrema that pervade measured rotor signals, while graph-based descriptions rarely move beyond pairwise edges even though a fault state is naturally a relation among many interacting points. This paper builds an interpretable, training-free hypergraph framework around three constructions. First, an NI-Louvain construction turns a recurrence graph into a hypergraph through overlapping communities, with each community becoming one hyperedge so that the higher-order coupling of the series is represented explicitly; an ablation shows this step carries the separability, which falls from about 0.59 to about 0.45 (means over 20 random seeds) when it is removed. Second, a variational mode decomposition multi-resolution framework runs the chain per intrinsic mode function, yielding a combined descriptor of 0.81 against 0.62 for the best single level. Third, a saturated symbolic entropy estimator smooths the entropy curve with a low-pass filter and provides two estimates from the same curve: the shoulder at the rapid-to-saturated junction; a stable estimation point; and the saturation time constant (τc), a front-end delay scale. It is not the more accurate estimator on the clean benchmark but the more stable and noise-robust one; a controlled sweep shows the downstream separability is strongly peaked at the small-lag recurrence window, and the role of the τc estimate is to land reliably inside that window rather than to read an accurate delay, so the saturated shoulder rule is valued for its stability and robustness. Validated on five chaotic systems, 180 measured rotor signals covering the four measured states, and the public MFPT bearing set, the framework is a training-free, interpretable alternative to learned classifiers on small rotor record sets.
No takes yet. Share an insight, caveat, or question.
Zhao et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: