Analysis of Plume Motion Characteristics in Selective Laser Melting Forming Process Based on Image Semantic Segmentation and Graph Convolution
Plume is one of the important dynamic characteristics in the process of selective laser melting.How to use plume information to study the process of selective laser melting is one of the directions that has been widely concerned in the monitoring of selective laser melting process.In order to better describe the plume information,Deeplabv3+network is used to conduct semantic segmentation of the image to obtain the plume contour,and establish a one-dimensional plume morphological feature.At the same time,combining with the time sequence signal,a graphic structure of plume motion feature is proposed,and the melting state category is obtained by clustering.In this way,the change of plume state during processing is discussed.Graph convolution was used to encode the dynamic features of the plume to realize the feature establishment and melting state recognition.The application of graph convolution in plume feature extraction was analyzed with the output of image convolution representation features.Based on the graph convolution,the graph convolution autoencoder and classifier are built to realize the classification of melting states,and the interpretability analysis of the graph convolution is carried out.The results show that the proposed plume features have high application value in the recognition of melting states.Finally,the recognition accuracy of the five types of melting states reaches 81.52％.The interpretability analysis shows how the features of the plume characterize the melting state,which provides a new idea for the study of the plume in the selective laser melting process.
NETL
NSTL
万方数据
