An edge-labeling graph neural network method for tool wear condition monitoring using wear image with small samples

Abstract

Tool wear condition monitoring (TCM) is of great significance to ensure manufacturing quality in milling processes, and the development of deep learning (DL) in recent years has led to increasing interest in DL-based TCM methods. However, most of these DL-based methods rely on large training samples to achieve good performances, which is expensive. In this paper, a new TCM method based on an edge-labeling graph neural network (EGNN) is proposed for small training datasets. First, the tool wear image is input into a convolution neural network (CNN) to extract features and obtain the features of the training samples. A fully connected graph is established based on these features, and the values of the edge labels are obtained by updating the nodes and edge features in the fully connected graph. Finally, the tool wear condition is predicted through the sample label of the support set and the predicted value of the edge connected with the query sample using a weighted voting method. The effectiveness of the proposed EGNN-based TCM method was demonstrated by its application to milling TCM experiments, and the results indicated that the proposed method outperformed three state-of-the-art methods (CNN, AlexNet, and ResNet) with small samples.