Online milling chatter identification using adaptive Hankel low-rank decomposition

Abstract

Regenerative chatter is a form of self-excited vibration that has been widely observed in high-speed milling operations and may result in poor part quality as well as limited process productivity. Towards intelligent machining, online chatter identification takes an essential place in the mitigation of chatter and thus has attracted much attention. Nevertheless, few works consider the underlying characteristics of the dynamics-based chatter prediction schemes, from which the signal-based chatter identification scheme may be crafted. Ignoring such connections may reduce the efficiency and sensitivity of the chatter identification system. To address this gap, this paper firstly clarifies the eigen-based characteristics of the dynamics-based stability prediction methods and reveals two transferable factors that should be concerned when processing milling signals. Then, a subspace-based detection-oriented signal decomposition method, i.e., the adaptive Hankel low-rank decomposition (AHLRD), is developed which can adaptively separate the chatter-related components from the observations in an efficient way. Afterward, two chatter indicators with physical significance are introduced to characterize the milling status from both the time and frequency domains. By incorporating with the support vector machine (SVM) predictor, the regenerative milling chatter can be automatically identified. The feasibility and effectiveness of both the AHLRD method and indicators are verified using dynamical simulation examples. A series of cutting experiments including both the secondary-Hopf and period-2 chatter cases are performed to verify the presented chatter identification method, while the vantages in accuracy, timeliness, and sensitivity are comprehensively verified through the comparison with the state-of-the-art methods.