Effect of different features to drill-wear prediction with back propagation neural network

• Thrust and torque are converted into equivalent thrust force and principal force.
• New features are generated from the resultant force of the equivalent forces.
• Features extracted from time domain and frequency domain are involved and compared.
• Wavelet packet transform decomposition and reconstruction is used to extract features.
• Importance of different input features is compared with an importance ranking algorithm.

In this paper, a back propagation neural network (BPNN) has been applied to predict the corner wear of a high speed steel (HSS) drill bit for drilling on different workpiece materials. Specially defined static and dynamic features extracted by a wavelet packet transform (WPT) from the resultant force converted from thrust and torque together with the cutting conditions (workpiece material, spindle speed, drill diameter, feed rate) are used as inputs to train the network to obtain a better output, drill corner wear. Drilling experiments have been carried out over a wide range and, features newly defined and conventional ones, features extracted from different frequency bands are compared.