AI-based sensorless tool chipping detection in milling using an Industrial Edge device

Abstract

The milling of difficult-to-machine materials, such as titanium alloys, is highly susceptible to tool failure mechanisms, including cutting-edge chipping and catastrophic tool breakage. Reliable and timely detection of cutting-edge chipping remains a persistent challenge, as undetected damage can lead to premature tool failure, dimensional inaccuracies, and degradation of surface integrity. In the present study, milling experiments were performed to investigate the feasibility of tool chipping detection using both laboratory-grade and industrial sensing systems. Cutting force and torque data were acquired simultaneously via a rotary-type dynamometer and a high-frequency (500 Hz) Industrial Edge device. Ground-truth measurements of tool chipping were obtained through post-process inspection using a stereomicroscope. Frequency-domain analysis revealed distinct spectral variations in the Fast Fourier Transform (FFT) of spindle current and torque signals obtained from the Edge device before and after tool chipping. Based on these observations, a Convolutional Long Short-Term Memory (Conv-LSTM) model was developed for automated tool chipping detection under varying cutting conditions. The model was trained and validated using frequency-domain features derived from both dynamometer and Edge device data. The proposed approach achieved average detection accuracies of 95% and 96% for the dynamometer and Edge device data, respectively. The key novelties of this work include a validated sensorless tool chipping detection framework using machine-internal Edge device signals, the integration of low-frequency industrial data with deep learning for reliable fault recognition, and a direct experimental comparison between laboratory-grade and deployable industrial sensing systems. The results demonstrate that data-driven analysis of process signals enables robust and reliable detection of cutting-edge chipping in milling operations. Moreover, the high detection performance achieved using Edge device data highlights the practicality of deploying the proposed method for real-time monitoring in smart manufacturing environments.