A digital apprentice for chatter detection in machining via human–machine interaction

Regenerative chatter in machining operations such as milling is a common process anomaly that limits productivity and part quality, which in turn lead to increased manufacturing costs. The industrial relevance of the problem has sparked many research efforts over the recent decades, with a growing interest in real-time chatter detection and suppression. Inspired by learning from human demonstration frameworks, this paper proposes a new approach to milling chatter detection via effective human–machine interaction, which facilitates knowledge transfer from an experienced machine tool operator to a “Digital Apprentice.” The proposed chatter detection approach acquires chatter-specific knowledge through a learnable skill primitive (LSP) algorithm designed to establish a robust chatter detection threshold from few-shot real-time demonstrations by an experienced human operator. In this work, digital audio data were acquired from milling experiments through a microphone mounted inside the milling machine. During the training phase, data for the human operator’s natural reaction to chatter were collected via a specially designed human–machine interface. The learned chatter detection thresholds were obtained via the LSP algorithm by temporally mapping the reaction time data to the audio signal. During the testing phase, experiments were conducted to validate the detection accuracy and detection speed of the learned chatter detection thresholds under different cutting conditions. The experimental validation results of the learned thresholds indicate an average chatter detection accuracy of 94.4%, with 55.6% of chatter cases detected before chatter marks are produced on a 4140 Steel workpiece, thus demonstrating the effectiveness of human–machine interaction in chatter detection.