Abstract: A tool breakage detection method was proposed based on the singularity analysis of vibration signal and self-learning support vector machine. The measured vibration signals were decomposed by the continuous wavelet transform， the wavelet modulus maxima (WTMM) and the Lipschitz index (Lips) were calculated. The state of tool breakage was recognized by Lips， and the Lips probability density function was fitted， which obeys the normal distribution. The support vector machine identification model of tool state was established by the parameter optimization of genetic algorithm based on mean value and variance of Lips (also called the optimal model). The tool breakage state was predicted by using this model， of which prediction accuracy increased gradually from 84% to 90%， and the robust of system prediction model was improved. Finally， the effectiveness of this method was verified by the experiments.

Key words: tool breakage, micro milling, wavelet singularity, support vector machine, self-learning