Bivariate empirical mode decomposition of grinding chatter signals

Abstract

Large numbers of experiments have shown that grinding chatter is one of the major forms of host fault performance in grinding processes. In view of this, more advanced monitoring techniques are required to ensure the high reliability of grinders. The empirical mode decomposition (EMD) technique has shown promise for meeting this requirement. In general, EMD has been limited to processing one-dimensional signals and is unable to deliver the information fusion function required for reliable chatter detection. In this paper, a bivariate EMD (BEMD) was assessed as a grinding condition monitoring technique. Conventional EMD and BEMD were compared by using them to process a simulated chatter signal. The BEMD technique showed a more powerful capability to process non-stationary and non-linear chatter signals. Moreover, BEMD was more effective for extracting features from multiple signals and detecting the phase information of intrinsic mode functions. The instantaneous energy, peak to peak, standard deviation and kurtosis parameters of the signal were able to be used as chatter feature vectors to describe the different vibratory states encountered during grinding. These feature vectors exhibit distinctive behaviors and could be applied as detectors of early grinding chatter.