Abstract
When bearing failure occurs, the fault information in the vibration signals is characterized by weakness, complexity, and a compound nature, making it difficult to accurately identify the compound fault of the bearing. 1D local binary patterns (1D-LBP) depict fault information of bearings from the aspect of local analysis. Unexpectedly, the quantitative 1D-LBP shows no consideration of the features of signals caused by bearing failure. Therefore, it is difficult to correctly quantize signals and extract failure information. The root mean square, an effective indicator of impact signatures in bearing faults, serves as the quantitative criterion for enhancing the 1D-LBP approach. Furthermore, signal fusion is applied to improve 1D-LBP, and quantized reconstructed signals are blended based on the harmonic fusion vector bispectrum (HFVB) to enhance failure features. Finally, the extraction of failure features and identification of faults in bearings are carried out through spectral analysis. The result of the comparison indicates that the proposed improved 1D-ILBP-HFVB method can extract feature information more accurately during bearing faults while effectively suppressing noise and distinguishing compound fault types in rolling bearings.