The life of microdrills (with radii less than 1 mm) is short and unpredictable. Estimation of tool wear and the pre-failure phase is more difficult than in conventional drilling operations since most special machine tools for microdrilling have stepping motors and create fluctuating force and vibration signals. In this paper, a new method is proposed for detection of the pre-failure phase of microdrills, just 0.2 to 1 second before the breakage occurs. The system measures the thrust force and the microdrill velocity of using a dynamometer and a laser vibrometer, respectively. The seven characteristic features of the signals are obtained by describing their shape and spike patterns. Adaptive Resonance Theory (ART2)-based neural networks are used for interpreting the features of the signals. The proposed system accurately classified all of the cases studied when a vigilance parameter of 0.995 was selected for the ART2.
