Backpropagation neural networks (BPNs) were used for on-line detection
of drill wear. The neural network consisted of three layers: input, h
idden, and output. The input vector comprised drill size, feed rate, s
pindle speed, and eight features obtained by processing the thrust and
torque signals. The output was the drill wear state which was either
usable or failure. Drilling experiments with various drill sizes, feed
rates and spindle speeds were carried out. The learning process was p
erformed effectively by utilising backpropagation with smoothing and a
n activation function slope. The on-line detection of drill wear state
s using BPNs achieved 100% reliability even when the drill size, feed
rate and spindle speed were changed. In other words, the developed on-
line drill wear detection systems have very high robustness and hence
can be used in very complex production environments, such as flexible
manufacturing systems. (C) 1998 Academic Press.