摘要
现有的研究中羽流已被证明与选区激光熔融加工过程有着密切关系且影响了最终成形质量,如何利用羽流等监测信息揭示选区激光熔融成形过程中熔融状态,是选区激光熔融成形过程监测被广泛关注的重要原因之一。为了更加完善地描述羽流信息,采用Deeplabv3+网络对图像进行语义分割,并使用标记图法建立一维的羽流形态特征;同时结合时序信号,构建图卷积自编码器与分类器,提出一种基于图结构的羽流运动特征,并进行聚类,得到熔融状态分类,讨论加工过程中羽流运动特征和熔融状态的关系。采用图卷积编码羽流动态特征,实现特征建立与熔融状态识别,并通过输出图卷积表征值,对图卷积在羽流特征提取中的应用进行可解释性分析。结果表明,提出的羽流特征能够实现五类熔融状态的识别且准确率达到了81.52%,并通过可解释性分析了羽流动态特征与熔融状态的内在关系,为选区激光熔融成形过程中监测信号与成形质量映射关系研究提供了新的思路。
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.
作者
林昕
毛杨坤
傅盈西
朱锟鹏
LIN Xin;MAO Yangkun;FUH Ying Hsi Jerry;ZHU Kunpeng(Precision Manufacturing Institute,Wuhan University of Science and Technology,Wuhan 430081;Key Laboratory of Metallurgical Equipment and Control Technology,Ministry of Education,Wuhan University of Science and Technology,Wuhan 430081;Institute of Intelligent Machines,Hefei Institutes of Physical Science,Chinese Academy of Sciences,Hefei 230031;National University of Singapore(Suzhou)Research Institute,Suzhou 215123)
出处
《机械工程学报》
EI
CAS
CSCD
北大核心
2024年第12期168-182,共15页
Journal of Mechanical Engineering
基金
国家自然科学基金资助项目(52175481)。
