基于AE并融合GMM与K-means的无监督颤振监测研究  

Unsupervised Chatter Detection Based on AE Merging GMM and K-means

作　　者：王丹[1] 张凤南 马岩尉 刘博[2] Wang Dan;Zhang Fengnan;Ma Yanwei;Liu Bo(Shenyang Institute of Technology,Shenyang 113122,China;不详)

机构地区：[1]沈阳工学院,沈阳市113122 [2]辽宁石油化工大学

出　　处：《工具技术》2025年第2期139-145,共7页Tool Engineering

摘　　要：金属切削过程中颤振的监测方法大致可分为颤振特征提取和聚类分析,其中提取方法有一定的局限性。本文提出一种基于大量未标记动态信号的无监督铣削颤振监测方法,该方法不依赖加工参数和环境,不需要标签,稳定性强,切削力信号来自多次铣削实验。该方法基于自动编码将信号的每一段压缩成二维,使用基于高斯混合模型和K-means合并的混合聚类方法对压缩信号进行聚类。所提出的方法在所有6个典型的无监督评价指标中都优于高斯混合模型和K-means算法。Chatter is prone to occur in the process of metal cutting,its detection methods can be roughly divided into extraction of chatter features and cluster analysis,but they have certain limitations.This paper proposes an unsupervised method for detection the states of milling based on a large number of unlabeled dynamic signals.The method does not depend on processing parameters and environment,does not require labels,and has strong stability.The cutting force signal comes from multiple milling experiments.The method is based on the auto-encode to compress each segment of the signal into a two-dimensional.Then,the compressed signals are clustered using a hybrid clustering method based on gaussian mixture model and K-means merging.The proposed method outperforms gaussian mixture model and K-means in all six typical unsupervised evaluation metrics.

关 键 词：颤振监测 高斯混合模型 K-MEANS 无监督聚类 自动编码器 

分 类 号：TG501[金属学及工艺—金属切削加工及机床] TH161.6[机械工程—机械制造及自动化]