Adaptive local binarization feature mode decomposition and its application in combined failure identification of rolling bearings

Abstract The combined failure of rolling bearings features weakness and complexity and is hard to recognize precisely. A 1D local binary pattern (1D-LBP) manifests failure information of rolling bearings from textural analysis. However, when signals are quantized with 1D-LBP, the periodic impact feature of fault signal itself will be excluded from consideration and consequently, the fault features will be hard to determine sufficiently. Feature mode decomposition (FMD) is sensitive to the impulse and periodicity of fault signals, but the number of decomposition modes and the length of the filter determines the accuracy of signal decomposition. To solve these problems, an adaptive local binarization FMD (ALBFMD) method is proposed. The ALBFMD method represents compound failure information of rolling bearings from textural feature extraction and inherent features of fault signals. Furthermore, with minimum permutation entropy as a criterion, the number of decomposition modes and the length of the filter of ALBFMD were adaptively determined. Based on the power spectrums of the reconstructed signals, the types of combined faults can be precisely identified. The proposed method is compared with FMD and the variational mode decomposition method and analyzed in different situations. Its superiority in terms of feature extraction and combined failure identification of bearings has been verified.