Selective Laser Melting Monitoring Based on the Plume and Its Motion Features

The monitoring of melting states based on in situ optical sensors has gradually attracted attention in the field of selective laser melting (SLM). It has been demonstrated that the plume is closely associated with the forming process and can significantly impact the final forming quality. To enhance the process descriptive capability of the plume, this study extracts a graph structure-based plume motion feature by integrating the temporal information of the plume. The k-means clustering algorithm is applied to excavate the hidden behavior patterns of the plume and reflect the nonstable forming processes. Subsequently, the graph convolutional network (GCN) is combined with the temporal gated-convolution (TCN) module to construct the melting state recognition model. This model achieves an accuracy of 84.33% in recognizing five classes of melting states. The interpretability of the model in extracting plume features is analyzed by the graph convolution characterization values, indicating hidden consistent features among similar melting states. This research establishes the mapping relationship between the plume behaviors and the melting state. As a result, it presents a novel method for feature-based intelligent monitoring of the SLM.