An Automatic Feature Extraction Method Based on Multiple Sensors

Abstract

Aiming at the problems of inaccurate monitoring and untimely fault detection in online tool condition monitoring system of CNC machine tools, an automatic feature extraction method based on multiple sensors is proposed. Firstly, different sensors are selected to collect vibration signal, three-phase current signal and acoustic emission signal during tool processing. Then the signals collected by all sensors are analyzed in time domain, frequency domain and wavelet domain respectively. After analyzing the signal, different features are extracted from it. For each feature, the least square method is used to obtain the fitting line. Finally, according to the comparison of the slope and square error of the fitting line, the characteristics that are highly correlated with tool wear are selected. These features are composed into an eigenvector to reflect the tool wear state. This method can monitor tool wear more accurately and timely.