An intelligent chatter detection method based on EEMD and feature selection with multi-channel vibration signals. (October 2018)

• Record Type: Journal Article
• Authors: Chen, Yun; Li, Huaizhong; Hou, Liang; Wang, Jun; Bu, Xiangjian
• Is Part Of: Measurement. Volume 127(2018)
• Page Start: 356
• Page End: 365
• Publication Date: 2018-10
• Subjects: Chatter detection -- Ensemble empirical mode decomposition -- Fisher discriminant ratio -- Support vector machine -- Multi-channel fusion -- Machining; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2018.06.006
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: A new chatter detection method using EEMD and feature selection is presented. EEMD highlights chatter characteristics in narrow decomposed frequency bands. FDR-based feature selection improves classification speed and performance. Fusion of two specific channel vibration signals gives better detection accuracy. Abstract: Chatter detection in metal machining is important to ensure good surface quality and avoid damage to the machine tool and workpiece. This paper presents an intelligent chatter detection method in a multi-channel monitoring system comprising vibration signals in three orthogonal directions. The method comprises three main steps: signal processing, feature extraction and selection, and classification. The ensemble empirical mode decomposition (EEMD) is used to decompose the raw signals into a set of intrinsic mode functions (IMFs) that represent different frequency bands. Features extracted from IMFs are ranked using the Fisher discriminant ratio (FDR) to identify the informative IMFs, and those features with higher FDRs are selected and presented to a support vector machine for classification. Single-channel strategies and multi-channel strategies are compared in low immersion milling of titanium alloy Ti6Al4V. The results demonstrate that the two-channel ( Ay, Az ) strategies based on signal processing and feature ranking/selection give the best performance in classification of the stable and unstable tests.