Tool wear estimation in turning by the use of non-stationary time series method

In the present paper, vibration signals recorded during a turning process are used for tool wear estimation. During the cutting process, tool acceleration signals are recorded with the use of an accelerometer installed on tool holder, for different wear values. Since the measured acceleration signals have time dependent variance and are non-stationary, time series with time dependent coefficients were used for modelling them and extraction of wear sensitive features. The extracted wear sensitive features are residual variance of different signals and are used as distance between the signals associated with worn tools and the baseline model. Based on analysis of distance between worn tools’ acceleration signals in different cutting directions and the baseline signal, a correlation was found. By the use of this correlation, a criterion was obtained for detection of critical wear stage. Such that the distance curves for feed and main cutting directions possess a minimum in the vicinity of critical wear level. Investigation of results revealed that the curves obtained through this method are more accurate than the ones obtained from stationary modelling method. The results obtained here can be used in online real-time algorithms of tool wear estimation.