流形学习在机械故障诊断中的应用研究  

An Investigation of Applying Manifold Learning to Diagnose Machinery Faults

作　　者：王冠伟[1] 张春霞[2] 庄健[1] 于德弘[1] 

机构地区：[1]西安交通大学机械学院,西安710049 [2]西安交通大学数学与统计学院,西安710049

出　　处：《工程数学学报》2012年第4期593-599,共7页Chinese Journal of Engineering Mathematics

基　　金：国家自然科学基金(50705073);中央高校基本科研业务费专项基金;教育部博士学科点专项科研基金(20100201120048)~~

摘　　要：本文针对信号采集系统的特性对流形学习方法性能的影响尚不明确这一问题,采用理论分析和模拟实验的方法,研究了信号采样系统的非线性、零点漂移等特性对流形学习算法性能的影响.结果表明,当信号采样系统的特性保持相对稳定时,流形学习方法可以在一定程度上容忍系统存在的非线性和零点漂移效应.为了使流形学习算法达到较好的效果,在数据的搜集和预处理过程中,应使得数据容易重构到一个高维空间中且它们之间的相似性易于度量.从而,本文的研究结果为流形学习方法在机械故障诊断中的应用提供了一定的理论基础.Currently, it is still not clear how the characteristics of a signal sampling system affect the performance of a manifold learning technique. This paper investigates the influence of the nonlinearity and zero-drift of a signal sampling system on the performance of a manifold learning technique theo-retically and experimentally. Based on the obtained results, the following conclusions can be yielded. When the characteristics of the signal sampling system maintain stable relatively, manifold learning can tolerate the existence of systemic nonlinearity and zero-offset effect to a certain extent. In order to make a manifold learning algorithm achieve good performance, it requires that the collected data should be easily reconstructed into a high-dimensional space and the dissimilarity between them can be reasonably measured. Therefore, the research results in this paper lay a theoretical foundation to the application of manifold learning methods in machinery fault diagnosis.

关 键 词：故障诊断 流形学习 ISOMAP算法 信号采样系统 

分 类 号：TH457[机械工程—机械制造及自动化]