On line prediction of roll grinding chatter based on EMD component and wavelet packet energy entropy

Abstract

To address the issue of partial feature loss in the single processing method within the time-frequency domain for roll grinding chatter, a combined time-frequency domain method is proposed to process signal feature. An intelligent algorithm is used to achieve online prediction of roll grinding chatter. Firstly, the empirical mode decomposition (EMD) method is utilized to decompose the vibration sensor signals, extrating the intrinsic mode function (IMF) while removing "spurious components" to calculate time domain characteristics associated with roll grinding chatter. Then, wavelet packet energy entropy is used to solve the frequency band node energy entropy values of acoustic emission sensor signals, obtaining frequency domain features characterizing the roll grinding chatter. Finally, the time-frequency domain features after dimension reduction is substituted into the intelligent algorithm model for online prediction of the roller grinding process. The results show that the the LV-SVM model achieves an average classification accuracy of 92.75%, with an average response time of 0.776 5 s. This verifies the validity of EMD and LV-SVM based on wavelet packet energy entropy in the time-frequency domain for online prediction of roller grinding chatter.

Keywords

• roll grinding chatter
• emd decomposition
• intrinsic mode function (imf)
• wavelet energy entropy
• least squares support vector machine (ls-svm)