MULTIMILLING-INSERT WEAR ASSESSMENT USING NON-LINEAR VIRTUAL SENSOR, TIME-FREQUENCY DISTRIBUTION AND NEURAL NETWORKS

Abstract

The objective of this study is to establish a signal processing methodology that can infer the state of milling insert wear from translational vibration measured on the spindle housing of a milling machine. First, the tool wear signature in a translational vibration is accentuated by mapping the translational vibration into a torsional vibration using a previously identified non-linear relationship between the two, i.e. a virtual sensor. Second, a time-frequency distribution, i.e. a Choi–Williams distribution, is calculated from the torsional vibration. Third, scattering matrices and orthogonalisation are employed to identify the time–frequency components that are best correlated to the state of wear. Fourth, a neural network is trained to estimate the extent of wear from these critical time frequency components. The combination of the virtual sensor, time–frequency analysis and neural network is then validated with data obtained from real cutting tests.