Abstract: Aiming at the problem that rolling bearing early fault features are too weak to identify the rolling bearing operation state quickly and effectively,a fault diagnosis method based on orthogonal neighborhood preserving embedding and relevance vector machine for roller bearing early fault identification was proposed.The multi-domain feature set which indicated the roller bearing early fault state was constructed.The low dimensional sensitive fault features were extracted from this multi-domain feature set based on the linear manifold learning Orthogonal Neighbourhood Preserving Embedding (ONPE) for dimension reduction,and the acquired low dimensional sensitive features were input to multi-kernel Relevance Vector Machine (RVM) to identify the early fault patterns.The roller bearing early fault experiment result showed that the proposed method had good performance in dimension reduction for high dimensional complex feature set,while the diagnosis accuracy was also better than single kernel RVM.

Key words: orthogonal neighborhood preserving embedding, ulti-kernel relevance vector machine, roller bearing, early fault, fault diagnosis

摘要： 针对滚动轴承早期故障特征微弱难以快速有效辨识的问题,提出一种基于正交邻域保持嵌入(ONPE)与多核相关向量机(RVM)的滚动轴承早期故障诊断方法。首先基于多域量化特征构造表征滚动轴承早期故障的多域特征向量,基于ONPE线性流形学习对多域特征向量进行约简降维处理,获取最能反映滚动轴承早期故障运行状态变化的低维敏感特征,随后将获取的低维敏感特征输入给多核RVM进行早期故障模式的分类辨识。通过分析滚动轴承早期故障的模拟实验数据表明,该方法对高维复杂的非线性早期故障特征具有良好的约简降维性能,而且比单一核函数RVM具有更好的诊断精度。

关键词: 正交邻域保持嵌入, 多核相关向量机, 滚动轴承, 早期故障, 故障诊断