CBM can avoid the occurrence of insufficient and excessive repairing .For the purpose of quantitative identification of bearing fault severity underlying CBM ,a GMM based approach was for‐mulated .The GMM was firstly trained by the multi-domain features including both time and frequen‐cy domain which were extracted from bearing fault-free stage .Subsequent feature vectors were then input to the trained model to compare their similarity degrees to the feature vectors of normal condi‐tions .Such similarity degree served as a fault severity index which was herein termed MDLLP(multi-domain log likelihood probability) .MDLLP had a upper limit of 1 ,which facilitated the determination of performance degradation levels in the field .Experimental results show that the proposed method and index are able to detect incipient bearing faults and can trend the fault progression well .It is im‐plied that the selection of optimal feature subsets has a substantial impact on the effects of the pro‐posed MDLLP .%视情维修可避免维修不足与维修过剩等问题，滚动轴承性能退化程度量化评估是实现视情维修的基础。提取滚动轴承早期无故障振动信号的时域、频域特征构建多域特征矢量，建立无故障轴承高斯混合模型（GMM ）。将轴承后期振动信号的多域特征矢量输入该 GMM 模型，得到测试样本与无故障样本之间的量化相似程度，以此建立多域对数似然概率（MDLLP）值作为滚动轴承性能退化定量指标。 MDLLP 的取值上限为1，便于实际使用中确定轴承性能退化状态。轴承疲劳试验表明，该方法能及时发现轴承早期故障，并能很好地跟踪故障发展趋势，最优特征的选择与变换对评估效果具有较大影响。
展开▼
