Multi-source heterogeneous sensor information fusion framework for intelligent online chatter detection in different milling conditions

Online chatter detection is a critical technology in intelligent manufacturing systems, essential for ensuring high-quality and efficient milling operations. Although artificial intelligence models have been developed to automatically identify chatter, the accuracy improvement is limited by the use of single sensor signals. Therefore, a multi-source heterogeneous sensor information fusion framework is proposed for intelligent online chatter detection in this paper. To effectively mitigate noise and eliminate interference from milling parameters, a heterogeneous sensor signal processing strategy is proposed based on wavelet packet decomposition and successive variational mode decomposition. Next, a multi-source, multi-stage, and multi-scale spatial-temporal fusion attention network is proposed for extracting chatter features and achieving high-precision chatter detection. It is noteworthy that multi-source signals are fused at the feature level, and comprehensive chatter features are extracted through the multi-source information fusion module, the multi-stage spatial-temporal feature extraction and fusion module, and the multi-scale gated channel attention module. In milling experiments across different conditions, the chatter detection performance of the proposed framework is evaluated in three scenarios. The results indicate that this framework can provide more accurate and reliable detection results compared to other methods.