AI-based tool wear prediction with feature selection from sound signal analysis

With the advancement of Industry 4.0, there has been a growing demand for the automation and digitalization of manufacturing processes, including machining. One of the core elements of this evolution is tool wear monitoring. In automated production systems, the condition of tools greatly influences production efficiency, cutting stability, and the quality of machined surfaces. The present study proposes an effective tool condition monitoring system based on cutting sound signature analysis and a machine learning model for milling processes. In the proposed system, the correlation between the sound signal and the tool flank wear under various cutting conditions is investigated. First, the measured sound signals in the milling process are extracted into a series of intrinsic mode functions (IMFs) using the ensemble empirical mode decomposition (EEMD). Hilbert transform (HT) is then applied to each IMF to generate the respective instantaneous frequencies, and the most significant statistic features correlated to the tool wear are selected using the collinearity diagnostics. Finally, an artificial neural network (ANN) model is designed to estimate tool wear levels. Experimental results confirm that the developed approach maintains excellent accuracy in tool wear prediction across of various cutting conditions. Moreover, the proposed approach has the potential to be implemented in practical applications as a cost-effective method for tool condition monitoring.