Normalized feature energy-guided adaptive frequency band division method for wheelset bearing compound fault detection

Resonance demodulation technology effectively extracts bearing fault features and identifies fault types. Its key step is the accurate detection of the informative frequency band (IFB) caused by faults. For compound faults, multiple IFBs from different faults must be identified. Given the complexity of the train operating environment, IFBs cannot be predetermined, and fixed band division often fails to localize IFBs accurately. To address these issues, we propose a novel frequency band adaptive division method based on the normalized feature energy (NFE) indicator, which adaptively divides the frequency band according to the bearing vibration signals to ensure accurate IFB detection. First, the original signal is converted into a Fourier spectrum (FS), which is adaptively divided based on set window lengths and a division strategy. Multiple window lengths are set according to the rotational frequency, each providing a frequency band division result, forming a multilevel tower-shaped boundary distribution. Next, for each subband, the NFE indicator defined in the envelope spectrum is calculated, considering the roller sliding effect directly related to fault features. The subband with the maximum NFE value is selected as the IFB, allowing for demodulation envelope analysis to identify fault types. To extend the method’s applicability to variable-speed conditions, we preprocess the variable-speed signals by resampling in the angular domain. Finally, we test the proposed method with simulated constant/variable speed data and experimental data. The results show that the proposed method is robust against impulsive noise and accurately detects IFBs, verifying its effectiveness.