Feature Extraction with Discrete Wavelet Transform for Drill Wear Monitoring

Abstract

The dynamics of drilling processes presents chaotic and unsteady characteristics, which prevent deterministic description. Vibration signals obtained during the microdrilling process contain rich information reflecting tool and process conditions. Experiments described in this paper show that as drill wear develops and intensifies, the energy distribution of the vibration signal tends to shift towards the low-frequency range. Traditional frequency domain analysis through the fast Fourier transform is not able to capture such transitions with desirable accuracy since the process is highly non-stationary. We propose a new method that combines the discrete wavelet transform with statistical estimations of the signal energy distribution to extract features describing such energy shifts quantitatively. Through a multiresolution transformation, four feature parameters most sensitive to drill wear conditions are extracted. A tool wear index is proposed as a linear function of the extracted features, which also represents the severity of tool wear. The effectiveness of the proposed method is shown through a case study at the end.