Abstract: Morphological operators are divided into two categories： noise reduction operators and feature extraction operators. The reported morphological undecimated wavelets use the identical morphological operator in each decomposition level, but it is challenging to capture the characteristic information of a signal simply by using a noise reduction or feature extraction operator repeatedly. Therefore, a layered-operator morphological undecimated wavelet is proposed in the paper, and different morphological operators are employed for each level of decomposition. The proposed method is more targeted and flexible in bearing fault feature extraction through the combination of noise reduction and feature extraction operators and with clear physical significance and easy to interpret. According to the characteristics of wheelset-bearing compound faults, a local characteristic amplitude ratio principle is proposed to select the most sensitive scale for each type of fault from multiple filtering scales to separate each fault effectively. The wheelset-bearing compound fault vibration signals are collected on a test rig, and the proposed layered operator morphological wavelet is applied to process the measured data. The results show that the proposed method can effectively detect the compound faults of wheelset bearings. Compared with the reported morphological undecimated wavelets, the layered operator morphological wavelet presents a superior performance in identifying the compound faults.

Key words: morphological undecimated wavelet, morphological filtering, fault diagnosis, compound fault, wheelset bearing