Machine learning based assessment of audio signal features for chatter identification in high-speed micromilling of thin-walled TC4 alloy

The present work identifies the audio signal features that are necessary to detect the chatter onset during high speed micromilling of thin-walled TC4 structure. These features have been derived from the short time Fourier transform of the signal to capture the low magnitude vibration of the micromilling process. Radial micromilling experiments at three different radial depths of cut and four different microphone tilt-angles shows 60° tilt angle as the optimum tilt angle. Entropy, kurtosis and shape factor have been found to be the optimum features based on the principal component analysis. Radial Basis Function has been found to be best suitable kernel with the highest prediction accuracy of 98 % for all the cutting conditions. The optimum features and kernel have been used in support vector machine (SVM) for classification of machining conditions into stable and chatter dominant zone. The presence of chatter marks has been validated using the machined surfaces.