Detection of tool failure in end milling

The use of an unsupervised adaptive resonance theory neural network to process the resultant force spectrum for the detection of tool failure in end-milling operations is reported in the paper. In this approach, an in-process identification of varying cutting geometries and complex cutting dynamics is not required for tool failure detection. The unsupervised neural network can classify the features of the resultant force spectrum under varying cutting conditions. As a result, tool failure can be detected when tool-failure features start to appear in end-milling operations. Experimental results are presented to confirm the feasibility of the proposed detection system.