1998 Volume 64 Issue 617 Pages 342-347
The fully automated running of machine tools requires early and automatic detection and prevention of abnormal cutting. Therefore, the purpose of this study is to determine a new approach to detect some kinds of abnormal cutting like self-excited chatter and tool flank wear in turning operation. Using this approach, the occurrence of abnormal cutting is evaluated by a neural network using a spectrum map of cutting tool vibration. The following conclusions were reached as a result of assessing this approach on actual cutting tests. (i) Learning times, where the total RMS error of the neural network with three layers could be within 0.001, reached minimum when number of units in hidden layer was 20, and this number of 20 was also minimum number of enough units to transfer all information about patterns from input layer to output layer. (ii) Capability of this approach to detect self-excited chatter was not affected by presence of tool flank wear. (iii) Width of tool flank wear could be detected within 0.1 mm error because shape of spectrum without self-excited chatter changed monotonously with increase of tool flank wear.