An Optimized VMD Method for Predicting Milling Cutter Wear Using Vibration Signal

Abstract

Tool wear has a negative impact on machining quality and efficiency. As for the nonlinear and non-stationary characteristics of vibration signals and strong background noises during the milling process, an identification method of the milling cutter wear state based on the optimized Variational Mode Decomposition (VMD) was proposed, in which the objective function is to minimize the Envelope Entropy (Ep); the various modes of the vibration signal are decomposed using the self-adaptive optimization parameters with Differential Evolution (DE). According to the cross-correlation coefficient in the frequency domain between Intrinsic Mode Function (IMF) and the original signals, the informative IMF components were selected as the sensitive IMF components to superimpose the reconstruction signal and extract the eigenvalues. The mapping relationship between the eigenvalues and the milling cutter wear degree is established by the Naive Bayes classifier method. The experimental results under the various operation conditions indicate that the proposed optimized VMD method possesses an excellent generalization performance. Compared with Empirical Mode Decomposition (EMD) and Ensemble Empirical Mode Decomposition (EEMD), it has better denoising capacity, and so can improve the identification accuracy of the milling cutter wear. Therefore, the processing quality and production efficiency are ensured effectively.

Tool condition monitoring for cavity milling based on bispectrum analysis and Bayesian optimized SVM

“…Babouri et al [14] reported that the energy and the mean power of the first Intrinsic Modes Functions (IMF) of the EMD shows great potential in tool wear status recognition and the method is validated by a turning process. Chang et al [15] developed a tool wear status identification method based on the optimize VMD, and the informative IMF was selected as the sensitive IMF components, in which the relationship between features and tool wear degree was built by the Naïve Bayes classifier method. Although the signal process and feature extraction methods mention above are proven to be effective and widely used, the phase information in signals is ignored, so higher order spectrum (HOS) is applied in this paper.…”