Tool Condition Monitoring using Competitive Neural Network and Hilbert-Huang Transform

Abstract

One of the major problems in fully automated manufacturing systems is the breakage and deterioration of the tools. Efficient tool condition monitoring systems are required to address such problem. In this study, a new method is proposed for tool condition monitoring for turning operation. The proposed method monitors the condition of the tool flank wear by classifying the tool into any one of the three states; initial wear, medium wear and severe wear. This classifying is done by a trained competitive neural network. The network is trained by using the instantaneous frequencies and amplitudes extracted from the audible emitted tool sound signal by using the new signal processing technique Hilbert-Huang transform. The proposed new method is tested by the audible sound signals collected from a turning machine while machining carbon steel with new, slightly worn and severely worn carbide inserts coated with Aluminum titanium nitride. From the marginal spectrum of Hilbert-Huang Transform analysis it is found that the amplitude of the emitted sound is increasing staidly as the tool flank wear is progressing with time. This correlation between the amplitude of the tool sound and tool flank wear enabled the trained competitive neural network to perform tool wear classification with 80% of accuracy. Hence, the new method can be implemented in tool condition monitoring of turning machines.