Abstract: To realize the automation and high accuracy of looseness extent diagnosis for connecting bolts of fan foundation,and to solve the key problems of looseness extent feature extraction,sensitivity feature selection,nonlinear dimension reduction and looseness extent stability identification,a looseness extent diagnosis method for connecting bolts of fan pedestal based on vibration sensitive feature and manifold learning dimension reduction was proposed.The time-frequency features were extracted from vibration signal to construct the origin looseness extent feature set,and the quantitative characterization for looseness extent of connecting bolts was realized.The calculation method of looseness extent sensitivity index was designed,and the sensitive features were selected to construct the looseness extent sensitive feature set that had stronger characterization capabilities.Moreover,the manifold learning method called Orthogonal Locality Preserving Projection (OLPP) was introduced to compress the looseness extent sensitive feature set into the low-dimensional looseness extent feature set,and the different looseness extents of connecting bolts was recognized by Weight K Nearest Neighbor Classifier (WKNNC).The feasibility and validity of the proposed method were verified by experimental results.

Key words: fan pedestal, looseness extent, sensitive feature, orthogonal locality preserving projection, weight K nearest neighbor classifier, fault diagnosis

摘要： 为实现在役风机基座螺栓松动程度诊断的自动化与高精度,解决松动特征提取与敏感特征选取、高维非线性约简与松动程度稳定识别的问题,提出基于振动敏感特征与流形学习约简的风机基座螺栓松动程度诊断方法。首先,融合振动信号时频特征构建出原始松动程度特征集,实现对松动程度的定量表征;设计出风机基座连接螺栓松动程度敏感性指标算法,选取敏感特征构建松动程度敏感特征集,增强特征集的表征性能。再应用正交局部保持映射流形学习算法对松动程度敏感特征集进行非线性约简,滤除冗余信息获得分类特性好的低维松动程度特征集,并输入加权最近邻分类器进行松动程度识别。工程应用结果证明了所提方法的可行性和有效性。

关键词: 风机基座, 松动程度, 敏感特征, 正交局部保持映射, 加权k最近邻分类器, 故障诊断