Acoustic signal-based chatter detection in milling using variational mode decomposition and k-NN classifiers

Abstract Chatter is a persistent challenge in milling operations, leading to reduced surface quality, tool wear, and productivity loss. This study proposes a novel chatter detection framework that integrates acoustic signal analysis with Variational Mode Decomposition (VMD) and advanced statistical feature extraction. A series of 44 end-milling experiments were undertaken on Al 7075 using an HSS four-flute cutter under varying spindle speeds (8.33–50 rps), feed rates (20–120 mm.min −1 ), and depths of cut (0.3–1.8 mm). The reconstructed acoustic signals were analyzed to extract 16 time- and frequency-domain indicators, with Spectral Entropy identified as the most discriminative feature for chatter detection. The Synthetic Minority Over-Sampling Technique (SMOTE) rectified dataset imbalance by expanding the dataset from 44 to 180 samples, improving classifier generalization. Five KNN-based models were evaluated, achieving up to 85% accuracy in distinguishing stable, transient, and unstable machining states. Comparative validation against the Stability Lobe Diagram confirmed the reliability of the proposed method. The findings demonstrate the potential of acoustic signal-based monitoring for real-time chatter prediction, providing a scalable solution for intelligent machining and Industry 4.0 applications.