A novel embedded deep learning framework for real-time online chatter detection in robotic milling

Hexiang Zhou, Zhoulong Li, Qunlin Cheng, Liyuan Pan, Long Shi, Limin Zhu,
A novel embedded deep learning framework for real-time online chatter detection in robotic milling,
Mechanical Systems and Signal Processing,
Volume 241,
2025,
113399,
ISSN 0888-3270,
https://doi.org/10.1016/j.ymssp.2025.113399.
(https://www.sciencedirect.com/science/article/pii/S0888327025011008)
Abstract: Chatter is a common issue during robotic milling processes and significantly degrades the quality of machined workpieces. To improve productivity, online chatter detection has been widely investigated. However, the characteristics of chatter vary with robot postures and machining parameters, presenting challenges for traditional detection methods. To address these challenges, this paper proposes a novel lightweight deep learning framework for real-time online chatter identification in robotic milling. A multi-band feature generation method is introduced to produce robot-modal dependent time–frequency maps with irrelevant frequency bands effectively removed. The feature maps are then fed into a specially designed lightweight deep learning network, which incorporates lightweight convolution to ensure accurate and efficient chatter identification. Finally, a cost-effective online chatter detection system is established, where the proposed analysis algorithm and deep learning network are implemented and tested on a microcontroller device. This system enables signal acquisition, processing, and communication with an upper-level monitoring system. To facilitate real-time data collection and chatter detection, a sliding window strategy is employed to acquire vibration signals. Experimental results demonstrate the effectiveness of the proposed method for chatter identification across various robot postures and milling parameters. The designed model shows strong generalization capabilities, with minimal parameters and computational cost. Additionally, the monitoring system exhibits remarkable real-time performance, successfully detecting chatter at its early stages.
Keywords: Robotic milling; Online chatter detection; Deep learning; Lightweight convolution; Embedded system