Real-world high-speed rail bogie vibration signals combine mechanical oscillations, deterministic periodic components (e.g., gear-mesh harmonics and wheel-rail interactions), environmental noise, and transients, making fault features hard to isolate. Diagnosis is further hindered by manual hyperparameter tuning, weak impulsive signatures at low Signal-to-Noise Ratio (SNR), overlapping spectral-temporal content with mode mixing, and high computational cost on long multi-channel records. We propose a multi-source vibration demodulation framework–an optimized Maximum Correlated Kurtosis Deconvolution (MCKD) with vibration-weighted grading distribution and Latent Cyclic Pattern Discovery (LCPD). The method couples spectral-negentropy-driven adaptivity for Minimum Entropy Deconvolution (MED)/MCKD filter length and regularization with wavelet-packet multiscale decomposition guided by correlation-kurtosis band selection, and periodic dynamic windows with a Vibration Amplitude-based Grading and Weighting Distribution (VAGWD), enabling adaptive deconvolution, enhanced weak-impulse extraction, and cycle-synchronous separation and quantification without empirical tuning. The LCPD module exploits envelope-cepstral cues, squared-envelope autocorrelation, and cyclic spectral coherence to recover hidden or time-warped periodicities under compound-fault conditions. Experiments on bogie datasets show higher diagnostic accuracy, greater noise robustness, and improved computational efficiency than MED, Multipoint Optimal Minimum Entropy Deconvolution Adjusted (MOMEDA), and baseline MCKD, enabling reliable and efficient compound-fault diagnosis in high-speed rail bogies.
扫码关注我们
求助内容:
应助结果提醒方式:
