邻域自适应增量式PCA-LPP在齿轮箱故障诊断中的应用  

Gear fault diagnosis based on an adaptive neighborhood incremental PCA-LPP manifold learning algorithm

作　　者：邓士杰[1] 唐力伟[1] 张晓涛[1] 

机构地区：[1]军械工程学院火炮工程系,石家庄050003

出　　处：《振动与冲击》2017年第14期111-115,132,共6页Journal of Vibration and Shock

基　　金：国家自然科学基金资助项目(50775219);军队科研资助项目

摘　　要：针对流形学习算法的增量处理问题,提出一种邻域自适应增量式PCA-LPP流形学习算法,阐述了算法的基本原理以及增量样本处理方法。对新增样本的引入,首先根据已有样本对协方差矩阵和相似矩阵进行增量更新,而后结合已有样本降维结果对新增样本降维结果进行估计,最后采用子空间迭代法实现新旧样本降维结果的更新。采用齿轮箱故障信号特征向量对邻域自适应增量式PCA-LPP流形学习算法进行检验,结果表明,邻域自适应增量式PCA-LPP流形学习算法降维后特征具有良好的故障分类识别效果。Aiming at the incremental processing in manifold learning algorithms, an adaptive neighborhood incremental PCA-LPP manifold learning algorithm was presented. The incremental learning principle of the algorithm was introduced. For introducing the incremental sample data, the adjacency and covariance matrices were updated according to the incremental by using the existing sample. Then, the dimension reduction result of the incremental sample was estimated based on the dimension reduction result of the existing sample and updated matrix. Finally, the dimension reduction results of incremental and existing samples were updated by using the subspace iteration method. The adaptive neighborhood incremental PCA-LPP manifold learning algorithm was applied in the processing of gearbox fault signals. The dimension reduction results by the incremental learning are of very small error compared with the batch learning. The spatial aggregation of incremental samples is basically stable, and the fault identification rate is improved.

关 键 词：增量式学习 自适应 流形学习 故障诊断 

分 类 号：TH165[机械工程—机械制造及自动化] TN911.3[电子电信—通信与信息系统]