Early chatter detection based on logistic regression with time and frequency domain features

Chatter is a kind of unstable vibration in metal removal processes, which causes severe damage to the workpiece and machine tool. To detect cutting chatter at an early stage, we propose a new chatter identification method based on status classification with logistics regression (LR) in this paper. The classification is based on the feature vector that is composed of the autocorrelation and improved wavelet packet entropy (IWPE). The autocorrelation coefficient characterizes the change of signal periodicity in the time domain and shows high sensitivity to chatter occurrence. To enhance sensitivity to chatter gestation, the IWPE is presented based on the wavelet packet entropy (WPE), which is a chatter indicator in frequency and energy domain. The IWPE is verified experimentally to identify chatter in the premature stage earlier than the WPE with the lead time of 0.8 s. To evaluate the performance of the proposed system, the end-face turning of a thin-walled disc was performed with various cutting parameters. The experimental results manifested feasibility and efficiency of the proposed detection method after being trained by experimental samples.