基于GWO-VMD的涡轮叶片特征提取  

Feature extraction of turbine blade based on GWO-VMD

作　　者：冯驰[1] 唐林伊 FENG Chi;TANG Linyi(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学信息与通信工程学院,哈尔滨150001

出　　处：《哈尔滨商业大学学报（自然科学版）》2024年第4期387-396,共10页Journal of Harbin University of Commerce：Natural Sciences Edition

基　　金：国家自然科学基金(61805056)。

摘　　要：燃气轮机涡轮叶片是燃气轮机的重要组成部分,长期工作在高温、高压、高速的恶劣环境下,健康的涡轮叶片能够保证燃气轮机高效率地工作.温度是反映燃气轮机涡轮叶片状态的关键参数,在涡轮叶片发生涂层脱落、断裂等故障时,叶片表面温度会发生变化,为了衡量涡轮叶片的健康状态,采用了温度这一叶片质量的重要指标,对数据预处理并进行特征提取,将涡轮叶片特征聚类.基于GWO-VMD提取叶片温度信号经分解后各IMF分量的4种熵特征,使用PCA对提取到的叶片特征进行降维,将降维后的特征送入FCM聚类算法,通过SSE和轮廓系数得到最佳聚类个数,实现对叶片特征的聚类.聚类效果表明,该特征提取方法能够较好地区分不同涡轮叶片的健康状态,为涡轮叶片健康监测的研究提供了一定思路.Gas turbine blades were an important part of gas turbines.They worked in the harsh environment of high temperature,high pressure,and high speed for a long time,and healthy turbine blades could ensure the high efficiency of gas turbines.Temperature was a key parameter to reflect the state of the turbine blades of a gas turbine.When the turbine blades suffered from failures such as coating peeling and fracture,the surface temperature of the blades changed.In order to measure the health status of the turbine blades,blade temperature,an important indicator of blade quality was used.The data were preprocessed and feature extraction was performed,and the turbine blade features were clustered.Based on GWO-VMD,the four entropy features of each IMF component of the decomposed leaf temperature signal were extracted,and PCA was used to reduce the dimensionality of the extracted leaf features.The dimensionality-reduced features were sent to the FCM clustering algorithm,and the optimal number of clusters was obtained by the silhouette coefficient and SSE.The clustering of leaf features was realized.The clustering effect showed that the feature extraction method could better distinguish the health status of different turbine blades,which provided a certain idea for the research of turbine blade health monitoring.

关 键 词：涡轮叶片 特征提取 灰狼优化算法 变分模态分解 模糊C均值聚类 

分 类 号：TN70[电子电信—电路与系统]