自适应Hessian LLE在机械故障特征提取中的应用  

Adaptive Hessian LLE in mechanical fault feature extraction

作　　者：李城梁[1] 王仲生[1] 姜洪开[1] 布树辉[1] 刘贞报[1] 

机构地区：[1]西北工业大学航空学院,陕西西安710072

出　　处：《振动工程学报》2013年第5期758-763,共6页Journal of Vibration Engineering

基　　金：国家自然科学基金资助项目(51075330;50975231;61003137;61202185)

摘　　要：针对机械故障中高维特征提取难的问题,提出了一种自适应选择邻域的流形学习算法。该算法基于流形局部弯曲度估计切空间,能使所有样本点自适应地选择邻域。将自适应选择邻域算法应用到海森局部线性嵌入(HLLE)中,改进后的HLLE在邻域图构建方面更能保证局部线性度,从而保证了Hessian LLE的降维性能。最终将自适应HLLE应用于滚动轴承4种不同状态的故障特征提取中,从提取样本的低维特征与识别精度的结果表明,自适应HLLE算法能够在邻域选择上对参数的选取具有较强的鲁棒性,提取机械故障的低维特征更加准确。In response to mechanical fault in high dimensional feature extraction problem,this paper presents an adaptive choosing neighborhood of manifold learning algorithm.The algorithm based on manifold local curvature estimation of tangent space,can make of all sample points adaptive selection of neighborhood.Adaptive selection of neighborhood algorithm applied to the Hessian locally linear embedding （HLLE）,the improved HLLE in the neighborhood graph is constructed to ensure the local linearity,thus ensuring the Hessian LLE reduction performance.Eventually the adaptive HLLE is applied to the rolling bearing of four kinds of fault feature extraction,extracted from the sample of low-dimensional feature and recognition accuracy results show,adaptive HLLE algorithm in neighborhood selection on parameter selection has a stronger robustness,extracting mechanical fault low-dimensional feature more accurate.

关 键 词：故障诊断 流形学习 自适应选择邻域 故障特征提取 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]