Analyse spectrale singulière des signaux vibratoires et Machine Learning pour la surveillance d'usure d'outils

Abstract

This study explores the use of Machine Learning techniques for tool condition monitoring in metal machining. Pseudo-local Singular Spectrum Analysis (SSA) of vibration signals raised on the tool holder, coupled to a pass-band filter allowed the definition and the extraction of features which are very sensitive to the tool wear. These features are defined from sums of Fourier coefficients of the signals reconstructed by SSA and of their residues, in judiciously selected frequency bands. The rates of recognition of wear are very good and exceed 80%. This study highlights two important aspects: strong relevance of information in high frequency vibration components and benefits of the combination of SSA and pass-band filtering to get rid of useless components (noise). © 2008 AFM EDP Sciences.