Tool Condition Monitoring in Machining by Fuzzy Neural Networks

Abstract

The Multiple Principal Component (MPC) Fuzzy Neural Network for tool condition monitoring in machining under varying cutting conditions is proposed. This approach is based on three major components of “soft computation,” namely fuzzy logic, neural network, and probability reasoning. The MPC classification fuzzy neural networks were built through training with learning data obtained from cutting tests performed in a reasonable range of cutting conditions. Several sensors were used for monitoring feature selection. Force, vibration, and spindle motor power signals were fused in multiple principal component directions to give a highly sensitive feature space. The tool conditions considered in the monitoring tests included sharp tool, tool breakage, slight wear, medium wear, and severe wear. The results showed success rates of approximate 94 percent in self-classification tests (i.e., the same data samples were used for both learning and classification), 84 percent in tests performed using different records for classification than those used for learning under the same cutting conditions, and about 80 percent in tests performed using samples obtained at different cutting conditions for classification than those used for learning within the same range of cutting conditions. The MPC fuzzy neural network classification strategy performed better than back-propagation trained feed-forward neural networks in these tests.