Rub-Impact Fault Detection via Variational Adaptive Chirp Mode Decomposition

Abstract

Rub-impact fault is a typical and critical failure in high-speed rotating machinery. Accurate extraction of its instantaneous frequency (IF) signatures is essential for condition monitoring and early fault warning. Adaptive chirp mode decomposition (ACMD) has been widely used for nonstationary signal analysis and rub-impact feature extraction. However, the local IF updating mechanism lacks physical constraints, and fixed smoothing parameters cannot adapt to dynamic fault feature evolution, which limits decomposition accuracy and stability. To address these problems, a novel variational ACMD (V-ACMD) is proposed in this article. The original heuristic IF update is upgraded to a global constrained variational optimization framework, which combines data consistency and multiorder smoothness priors to improve the dynamic consistency and noise robustness of IF estimation. Meanwhile, a data-driven adaptive mechanism is designed for the regularization parameters, while a prior-guided and mode-adaptive strategy is developed for the bandwidth parameter. Regularization parameters are dynamically adjusted according to the energy ratio between the demodulated signal and the differential signal. Bandwidth parameters are optimized via a two-stage coarse-to-fine search strategy based on a quality factor. In addition, a periodic extension method is adopted to suppress the endpoint effect. Numerical simulations, rotor dynamic simulations, and experimental results demonstrate that the proposed V-ACMD method can achieve superior performance in high-precision IF tracking, strong noise robustness, and high-fidelity fault feature extraction for rotor rub-impact fault detection.