Online milling chatter detection via data-driven adaptive chirp mode decomposition and multiscale sample entropy

To address the challenges associated with the online identification of early milling chatter—where diagnostic signatures are weak and easily obscured by forced vibrations (e.g., spindle rotation and tooth passing components) and background noise—this paper proposes an online chatter monitoring method based on data-driven adaptive chirp mode decomposition (DD-ACMD) and multiscale sample entropy (MSE). First, DD-ACMD utilizes a data-driven estimation of the initial instantaneous frequency to perform adaptive decomposition, thereby enabling the robust extraction of intrinsic modal components under non-stationary conditions. Subsequently, based on modal frequency characteristics, components associated with forced vibrations are identified and eliminated to reconstruct a chatter-sensitive signal with an enhanced signal-to-noise ratio. The MSE of the reconstructed signal is then computed to formulate a chatter detection index; this index is further integrated with a sliding window and an adaptive threshold for real-time state determination. Experimental results from milling processes demonstrate that the proposed approach effectively separates chatter-related components from forced vibrations, maintains stable detection performance under strong noise, and improves the sensitivity and accuracy of early chatter recognition, thereby providing practical support for online early warning in milling operations.