Online detection of milling chatter based on multi-order graph convolutional neural network

Chatter is a type of self-excited vibration that occurs during machining, particularly when processing weak-rigidity aerospace parts, adversely affecting machining quality and efficiency. In order to prevent the detrimental effects of chatter in milling processes, a physically interpretable multi-order graph convolutional neural network (GCN)-based method is proposed for online chatter detection. In this method, the vertex and edge of the network are first constructed through the measured multi-channel vibration signal time series segments with short duration, establishing the groundwork for online chatter detection. Then, a multi-order GCN is developed by gathering feature information from multiple orders of neighbors of the target vertex to enhance its accuracy. Next, based on the multi-order GCN, a deep neural network is constructed and trained using the vibration data obtained by numerous milling experiments. During the training process, the edge weights are optimized by the gradient descent method. After that, an online chatter detection system is constructed, which demonstrates robust noise resistance through tests with different levels of noise interference added. Finally, experiments under various milling conditions validate that the proposed method can accurately detect milling chatter at its early weak stage, thereby successfully preventing its adverse effects.