Intelligent Feature Engineering for Monitoring Tool Health in Machining

This chapter explores the development of a data-driven digital twin to detect the health of cutting tools in precision machining processes. A set of cutting tool run-to-failure machining tests was conducted in a semi-production CNC milling machine by changing the cutting parameter settings. Real-time data was acquired from various sensors during the tests. The sensory data were processed to explore features that correlate with the deteriorating health of the cutting tool. Specifically, we processed the signals in the time–frequency domain using wavelets to derive informative features. Our analysis presents evidence that a pool of features exists in higher wavelet subspaces that are informative of the cutting tool condition. The variance and kurtosis of the acquired signals explain the changes to the cutting tool condition. A comparative study performed between time–frequency (wavelets) and other domains suggests benefits of feature engineering with time–frequency analyses. An intuitive explanation of the features informative of the tool condition is shown by exploring the frequency spectrum of the signals. A strong linear correlation of 0.97 was obtained between the chosen feature and tool wear. A supervised machine learning-based monitoring system is developed that exploits the variability of information in different wavelet subspaces to forecast the tool wear. Vibration and force signals are found to be the most informative sensors of tool wear progression. This chapter highlights the importance of feature engineering and a robust low-cost machine learning algorithm for tracking the tool condition.