Abstract: A bearing performance degradation assessment method based on Bayesian inference and Self-Organizing Map (SOM) was proposed. Independent Component Analysis (ICA) was used to extract feature set of normal operation bearing from original feature set, and a SOM model was established to describe the healthy bearing state space. Furthermore, Negative Log Likelihood Probability (NLLP)-based performance quantification index and Bayesian Inference-based failure probability calculation method were developed to evaluate and identify dynamical running states of bearings online. In comparison with traditional feature indexes and Support Vector Data Description (SVDD) based performance degradation assessment, the experimental results of a bearing test bed showed that the proposed evaluation index was more effective to quantify the performance degradation propagation on the whole life of bearings. It could provide important healthy information for making maintenance plan.

Key words: bearing, performance degradation assessment, independent component analysis, self-organizing map, Bayesian inference, fault diagnosis

摘要： 为在线识别和评估轴承的动态运行状态，提出一种基于贝叶斯推论和自组织映射的轴承性能退化评估方法。首先运用独立成分分析算法从原始特征集提取表征轴承正常运行的特征集，建立描述轴承健康状态的基准自组织映射模型，进而提出基于负对数似然概率的设备性能量化评估指标和基于贝叶斯推论的失效概率计算方法，在线识别和评估轴承的动态运行状态。通过在轴承全寿命测试床的实验结果表明，与一些传统的特征值指标和基于支持向量数据描述的性能退化评估方法相比，提出的评估指标可有效地量化轴承的全寿命性能退化过程，为进一步制定维护计划提供重要的设备健康信息。

关键词: 轴承, 性能退化评估, 独立成分分析, 自组织映射, 贝叶斯推论