摘要
针对旋转机械高维复杂故障特征数据难以快速准确辨识的问题,提出一种基于等距映射非线性流形学习与加权KNN(K-nearest neighbor)分类器相结合的旋转机械故障诊断方法。在由时域统计指标和内禀模态分量能量构造的原始特征空间中,首先利用等距映射非线性流形学习算法提取旋转机械故障状态变化的本质特征,随后将提取的低维本质特征输入给加权KNN进行旋转机械的故障模式辨识。通过对齿轮箱的实验数据分析表明,该方法不仅对高维复杂的非线性故障特征具有良好的降维性能,而且故障识别率较之传统方法也明显提高,能够有效识别出高维特征空间的非线性故障特征。
Aiming at the problem that rotating machinery high dimension complex fault features are difficult to be identified quickly and accurately, a rotating machinery fault diagnosis method based on isometric mapping nonlinear manifold learning and weighted K-nearest neighbor （KNN）classifier is proposed. In the original feature space constructed from time domain statistics and intrinsic mode energy component, the isometric mapping nonlinear manifold learning algorithm is used to extract the essence features of rotating machinery fault state variation and carry out non-linear multi-dimensionality reduction; then, the extracted low-dimensional fault features are transmitted to the weighted KNN classifier for rotating machinery fault mode identification. The analysis of the experimental data for a gearbox shows that this method not only has good dimensionality reduction performance for high-dimensional complex non-linear fault features, but also significantly improves the fault recognition accuracy compared with traditional method. This method can effectively identify the nonlinear fault features existing in high dimensional feature space.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2013年第1期215-220,共6页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(51275546)
重庆市自然科学杰出青年基金(CQcstc2011jjjq70001)资助项目
关键词
流形学习
等距映射
加权K近邻
旋转机械
故障诊断
manifold learning
isometric mapping
weighted K-nearest neighbor （ KNN ）
rotating machinery
fault diagnosis
