Abstract: Under the condition of variable rotational speed, the weak fault features of rolling bearings are easy to be flooded and affected by the strong noise, which leads to the difficulty of identifying the fault features. Based on this, a combination of Order Frequency Spectral Coherence (OFSC) domain and low-rank sparse optimisation Go-Decomposition (GoDec) is proposed as a fault feature extraction method. (Go-Decomposition, GoDec) for fault feature extraction. Firstly, the OFSC calculation of the fault signal is performed according to the cyclic smoothness property of the angular-time domain of the variable speed fault signal. Secondly, according to the sparsity and low-rank nature of the fault features and background noise in the order-frequency spectral correlation domain, the fault features in the order-frequency domain are extracted by introducing the kernel paradigm and L1 paradigm to optimise the GoDec method. Finally, in order to highlight the fault features, the sparse components are feature-enhanced using Enhanced Envelope Order Spectrum (EEOS). The performance of the proposed method is verified by variable speed simulation signals and experimental data, and quantitative analysis is carried out. The results show that the accuracy of fault feature recognition of the proposed method is better than that of single sparse constraint, original GoDec and order ratio analysis methods, which has obvious advantages.

Key words: rolling bearings, variable speed conditions, GoDec, order spectral correlation, low order sparse

摘要： 在变转速条件下，滚动轴承微弱故障特征易被强噪声淹没并影响，导致故障特征难以识别，基于此提出了一种阶频谱相关(Order Frequency Spectral Coherence, OFSC)域和低秩稀疏优化Go分解(Go-Decomposition, GoDec)相结合的故障特征提取方法。首先，根据变转速故障信号的角度-时间域循环平稳特性，对故障信号进行OFSC计算。其次，根据故障特征和背景噪声在阶频谱相关域的稀疏性和低秩性，通过引入核范数和L1范数优化GoDec方法，对阶频域中的故障特征进行提取。最后，为了凸显故障特征，利用增强包络阶次谱(Enhanced Envelope Order Spectrum, EEOS)对稀疏分量进行特征增强处理。通过变转速仿真信号和实验数据验证所提方法的性能，并进行量化分析。结果表明：所提方法故障特征识别准确度优于单稀疏约束、原始GoDec和阶比分析方法，具有明显优势。

关键词: 滚动轴承, 变转速工况, GoDec分解, 阶频谱相关, 低秩稀疏