Regenerative chatter identification based on fast SWD and mRMR-RF in robotic milling

Chatter is a significant challenge in robotic machining. By identifying chatter through monitoring signals, abnormalities in machining states can be detected rapidly, thereby preventing further damage. To address this issue, a fast swarm decomposition (FSWD) method was proposed, and the parameters for different frequencies swarm filtering (SWF) were obtained using non-dominated sorting whale optimization algorithm (NSWOA). To improve the extraction of chatter components, the primary frequency range of chatter was defined as the decomposition interval. To overcome the limitations of single-feature analysis, the maximum relevance minimum redundancy (mRMR) method was employed to select the optimal feature subset in time-frequency domain, and the machining state identification model was established based on random forest (RF). Finally, a robotic milling experiment of 5A06 aluminum alloy was conducted to validate the model. The results showed that the proposed model could accurately extract the intrinsic mode functions (IMFs) related to chatter. The identification accuracy for no-load, stable cutting, early regenerative chatter, and severe regenerative chatter reached 95%, while the computation time was less than 25 ms. Moreover, the mRMR-RF model’s accuracy was 5.8% higher than that of mRMR-KNN and 0.7% higher than that of mRMR-BPNN, highlighting its advantages.