基于小波包和拉普拉斯特征值映射的柱塞泵健康评估方法  

Health assessment for a piston pump based on WPD and LE

作　　者：王浩任[1] 黄亦翔[1] 赵帅[1] 刘成良[1] 王双园[1] 张大庆 

机构地区：[1]上海交通大学机械系统与振动国家重点实验室,上海200240 [2]山河智能装备集团,长沙410100

出　　处：《振动与冲击》2017年第22期45-50,共6页Journal of Vibration and Shock

基　　金：国家科技支撑计划(2014BAA04B01);国家自然科学基金(51305258);上海市科委项目(1411104600)

摘　　要：柱塞泵是液压系统的关键部件之一,监测其健康状态对液压系统的可靠运行具有重要意义。提出一种基于小波包和流形学习的方法,用于分析柱塞泵出口振动信号,从而对其进行健康评估;该方法利用小波包对原始信号进行分解,从中提取用于描述柱塞泵健康状态的有效特征群;把提取的高维特征群作为输入,利用并比较多种流形学习方法进行特征降维,选取状态识别准确率最高的拉普拉斯特征映射方法,建立起的特征向量到健康状态之间的对应关系,实现液压泵健康状态监测的分类要求。实验结果表明,采用小波包和拉普拉斯特征映射相结合的方法可以有效提高柱塞泵状态评估的准确性。A piston pump is one of the key components in a hydraulic system,monitoring its health condition is of significant importance for reliable performance of the hydraulic system. Therefore,a health state evaluation method was proposed based on the wavelet packet decomposition（ WPD） and the manifold learning by analyzing the vibration signals at the piston outlet. The wavelet packet method was used to decompose original signals and effectively extract health state features group from them. The high dimensional feature group was set as an input and multiple manifold learning methods were conducted and compared for dimensional reduction. The Laplacian eigenmaps（ LE） method of the highest accuracy was used to establish a relationship between feature vectors and health states,which achieved the aim of health state classification. It is shown that the combination of the wavelet packet decomposition and the manifold learning method improve the accuracy of piston pump health state evaluation.

关 键 词：小波包分析 流形学习 柱塞泵 拉普拉斯特征映射 健康状态评估 

分 类 号：TP306[自动化与计算机技术—计算机系统结构]