Timely online chatter detection in end milling process. (15th June 2016)

• Record Type: Journal Article
• Authors: Fu, Yang; Zhang, Yun; Zhou, Huamin; Li, Dequn; Liu, Hongqi; Qiao, Haiyu; Wang, Xiaoqiang
• Is Part Of: Mechanical systems and signal processing. Volume 75(2016)
• Page Start: 668
• Page End: 688
• Publication Date: 2016-06-15
• Subjects: Chatter detection -- Hilbert–Huang transform -- Energy aggregation -- Online; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2016.01.003
• Languages: English
• ISSNs: 0888-3270

Abstract

Abstract: Chatter is one of the most unexpected and uncontrollable phenomenon during the milling operation. It is very important to develop an effective monitoring method to identify the chatter as soon as possible, while existing methods still cannot detect it before the workpiece has been damaged. This paper proposes an energy aggregation characteristic-based Hilbert–Huang transform method for online chatter detection. The measured vibration signal is firstly decomposed into a series of intrinsic mode functions (IMFs) using ensemble empirical mode decomposition. Feature IMFs are then selected according to the majority energy rule. Subsequently Hilbert spectral analysis is applied on these feature IMFs to calculate the Hilbert time/frequency spectrum. Two indicators are proposed to quantify the spectrum and thresholds are automatically calculated using Gaussian mixed model. Milling experiments prove the proposed method to be effective in protecting the workpiece from severe chatter damage within acceptable time complexity. Highlights: Signal energy distribution transition when chatter develops is analyzed. The energy aggregation process is adopted as the basis of the detection method. A revised Hilbert-Huang transform is proposed to process the signal. Two indicators are proposed to quantify the spectrum. Thresholds are automatically determined using Gaussian mixed model. Comparisons are made with existing wavelet method. Experiments prove the method to be effective within acceptable time complexity.