一种融合型异常检测算法及其在轴承性能退化评估中的应用

摘要

用自回归模型(autoregressive model,AR)提取早期无故障滚动轴承的振动样本以及同型号同位置失效滚动轴承(简称同类轴承)的失效样本,用早期无故障样本和失效样本建立模糊C均值(Fuzzy C Mean,FCM)和隐马尔科夫(Hidden Markov Model,HMM)性能退化评估模型,然后得到正常指标和同类轴承的失效指标,把这两个指标作为输入特征建立FCM模型,待测数据通过保持模型不变连续迭代的方式输入模型中,描绘出性能退化曲线.该方法集中了空间统计距离和相似度方法两者的优势且不需要轴承失效数据.实验表明所提出的评估方法得到的评估指标能实时监测滚动轴承的性能退化趋势并且可以及时发现早期故障.%The paper extracts the early fault-free vibration signal characteristics and the failure characteristics of rolling bearing with the same type and position using autoregressive model.The FCM and HMM are established using the early failure-free samples and the failure samples.And then the normal index and failure index of similar bearings are obtained.The FCM model is established using the two indexes.The measured data are input into the model by keeping the model unchanged and iterating.And the performance degradation curve is drawn.The method combines the advantages of spatial statistical distance and similarity methods and does not require bearing failure data.Experimental analysis shows that the performance degradation method proposed in this paper can evaluate the performance degradation of rolling bearing in real time and can detect the early failure in time.