The service state of axle-box bearings in rail vehicles directly affects the safety of train operation. Current fault diagnosis methods based on spectral coherence (SCoh) often ignore the key information in the discrete fault bands excited by bearing faults, which reduces the diagnostic perform-ance for weak faults. To solve this problem, this paper proposes a fault diagnosis method based on multiband clustering weighted envelope spectrum (MCWES). First, the original SCoh is reconstructed, and the proposed autocorrelation period ratio (ACR) calibration algorithm is utilized to achieve automatic identification of potential fault frequencies within each spectral frequency slice (SFS). Subsequently, a weighting metric is constructed to evaluate the richness of fault information in each SFS to determine the initial central frequency band for the optimization process. Next, an equal-proportional extended optimization structure is designed to generate EPEOgrams for detecting the optimal demodulation bands in each initial center band. Finally, optimal demodulation bands (ODBs) with similar potential cyclic frequencies (PCFs) are clustered to construct MCWES for efficient detection of bearing faults. The results of axle-box bearing bench tests on rail vehicles show that the proposed method can fully integrate bearing fault information in different narrow frequency bands without relying on nominal fault cycle information. Compared with the existing state-of-the-art methods, the method identifies the axle-box bearing fault characteristics and their harmonics more effectively, which significantly improves the diagnostic performance of weak faults.

Axle-box bearing, Multiband clustered weighted envelope spectra, Spectral coherence, Band selection, Fault diagnosis.