关键词: 机器人铣削, 颤振辨识, 变分模态滤波, 注意力机制

Abstract: A method was proposed for identifying chatters in heavy-duty robotic milling systems by integrating variational mode filtering with fixed parameters, envelope filtering and an attention mechanism network identification. Initially, variational mode filtering theory was applied to eliminate non-chatter signal components in the high-frequency ranges by optimally selecting a quadratic penalty. Then, to swiftly identify the current machining conditions, the envelope filtering method was employed, leveraging signal time domain distribution and the frequency domain mapping law to remove the spindle speed-related signal components in the low-frequency ranges. Subsequently, a network identification model incorporating an attention mechanism was developed to identify preprocessed multi-temporal short-term signal segments for machining condition identification, followed by verification experiments on heavy-duty robotic milling systems. Experimental analysis results demonstrate that by eliminating non-chatter signals in the high-frequency ranges and spindle speed-related components in the low-frequency ranges, the accuracy of regenerative chatter identification is significantly enhanced, achieving an identification accuracy of 98.75%. Compared with alternative identification methods, the proposed method may effectively identify regenerative chatters during heavy-duty robotic milling processes, thus offering valuable technical support for future online chatter suppression of heavy-duty robotic milling.

Key words: robotic milling, chatter identification, variational mode filtering, attention mechanism