Milling Tool State Classification through Detailed Coefficients of Discrete Wavelet Transform as Feature Set – A Comparative Study

The tool state monitoring(TSM) is highly required to ensure the optimal tool usage without compromising the quality of surface finish it produces on the work piece. Vibration based monitoring is highly reliable to differentiate the tool state through proper signal processing techniques. In this study, the efficacy of Discrete Wavelet Transform (DWT) under different cutting conditions in face milling process is analyzed through deep learning. The mean squared energy (MSE) of the detailed coefficient of DWT decomposed at eight levels are computed and forms the feature set. The cutting inserts with different state such as Fresh Tool (G), Tool wear on the flank surface (FW), Tool wear on the rake face (FL), and broken tool (B) are considered in this study. The salient features are identified through Decision Tree (DT) algorithms and the classifying ability of Long Short Term Memory (LSTM) model is analyzed. The results are compared with the Fuzzy Logic (FL) classifier. The results show that the LSTM with MSE of detailed coefficients of bior 3.7 DWT feature outperformed the FL and exhibits excellent performance in classifying the tool state of face milling process.