Adaptive enhanced CYCBD fault diagnosis algorithm for bearings

Abstract The maximum second-order cyclostationary blind deconvolution (CYCBD) is a novel blind deconvolution algorithm that excels in recovering bearing fault pulses. However, several limitations have been identified in its application: the cyclic frequency set must be accurately pre-estimated; the ability of CYCBD to recover pulses improves with the length of the deconvolution filter, but excessive filter length can distort the signal and significantly increase computation time. To solve the above problems, the adaptive maximum second-order cyclostationary blind deconvolution (ACYCBD) algorithm that weighs the performance and efficiency of the algorithm is proposed in this paper. Firstly, an improved envelope harmonic product spectrum (IEHPS) method for cyclic frequency set estimation is proposed, with its effectiveness validated through simulations and experiments. Then, the performance efficiency ratio (PER) metric of the reweighted envelope spectral harmonic interference ratio (RESHIR) is proposed to adaptively select the filter length using an isobaric search strategy after considering the performance of the CYCBD and the combined cost of the runtime. Finally, an adaptive enhanced CYCBD algorithm, integrating IEHPS and RESHIR-PER, is developed. The effectiveness of this algorithm is validated using high-speed train traction motor bearing signals and the open bearing dataset from XJTU-SY. The results demonstrate that the proposed algorithm exhibits superior robustness and accuracy compared to SMHD, IMCKD, and ACYCBD algorithms.