TOOL-WEAR PREDICTION USING ARTIFICIAL NEURAL NETWORKS

A mixed-oxide ceramic cutting tool (type K090) has been used to machin e grey cast iron (grade G-14) in a turning process. Different values o f feed rate and cutting speed have been used for machining at a consta nt depth of cut. Tool life and failure mode have been recorded for eac h experiment and the associated data have been used to train an artifi cial neural network (multi-layer perceptron) using the back-propagatio n algorithm. The trained network has been used to predict tool lives a nd failure modes for experiments not used in training. The best result s are 58.3% correct tool-life prediction (within 20% of the actual too l life) and 87.5% correct failure-mode prediction, but it was felt tha t these could be improved significantly if more real data was generate d for the training of the neural network.