【作者】 周志红； 周新聪； 严新平； 袁成清；

【Author】 ZHOU Zhi-hong ZHOU Xin-cong YAN Xin-ping YUAN Cheng-qing (Reliability Engineering Institute,Wuhan University of Technology,Wuhan 430063,China

【机构】 武汉理工大学可靠性工程研究所；

【摘要】 在众多的磨粒特征参数中,存在着大量的噪音和冗余信息．为了提高磨粒自动识别系统的性能和效率,在进行磨粒分类之前必须对这些参数加以评估、优化,从中选择出最有代表性的特征来．本文在介绍特征抽取及特征选择原理的基础上,针对磨粒分析的特点,提出了4种磨粒特征的优化方法:非线性主成分法、 BP 神经网络、模拟退火法及遗传算法．更多还原

【Abstract】 There are much irrelevant or redundant information in wear debris feature parameters.In order to improve the performance and accuracy of automatic wear debris analysis system,it is important to evaluate and optimize these parameters.In this paper,the theory of feature extraction and selection is discussed.Four methods and models are also proposed to wear debris feature evaluation and optimization problems such as non -linear principal component analysis,BP neural network,simulated annealing and genetic algorithms.更多还原