基于超分辨率的选区激光熔融熔池形貌研究
Research on Melt Pool Morphology of Selective Laser Melting Based on Super-Resolution
吴磊 1孔令成 2傅盈西 3朱锟鹏2
作者信息
- 1. 中国科学院合肥物质科学研究院,安徽合肥 230031;中国科学技术大学研究生院科学岛分院,安徽合肥 230026
- 2. 中国科学院合肥物质科学研究院,安徽合肥 230031
- 3. 新加坡国立大学新国大苏州研究院,江苏苏州 215123
- 折叠
摘要
在选区激光熔融加工过程中,包含大量加工状态和工艺参数信息的熔池图像存在噪声污染、熔池边缘模糊等问题,因此提高熔池图像的分辨率对监测熔池状态具有重要意义.提出一种基于深度学习的熔池图像超分辨率重建算法,用于提取低分辨率熔池图像的浅层特征和深层特征并学习低分辨率图像到高分辨率图像之间的映射关系,使用亚像素卷积上采样完成熔池图像重建过程.分别对采集的高分辨率、退化后的低分辨率、重建的高分辨率熔池图像进行图像预处理和提取熔池形貌特征,以采集的高分辨率熔池图像的形貌特征为标签计算熔池特征误差率.结果显示,重建的熔池图像不仅可以提高分辨率,还大幅度降低了熔池形貌特征提取的误差率,为选区激光熔融成型过程监测和精准调控加工工艺参数提供了保证.
Abstract
In the process of selective laser melting,the melt pool images containing a large amount of processing states and process parameters information have problems such as noise pollution and blurred edge of the melt pool.Therefore,improving the resolution of the melt pool images is of great significance for monitoring the states of the melt pool.In this paper,a super-resolution reconstruction algorithm of melt pool images based on deep learning is proposed,which is used to extract shallow and deep features of low-resolution melt pool images and learn the mapping relationship between low-resolution images and high-resolution images.The melt pool images reconstruction processes are done using subpixel convolutional up-sampling.Perform image preprocessing on the collected high-resolution,degraded low-resolution,and reconstructed high-resolution melt pool ima-ges,respectively,and extract melt pool morphology features,using the morphology features of the collected high-resolution melt pool images as labels to calculate the error rate of the melt pool features.The results show that the reconstructed melt pool images can not only improve the resolution,but also greatly reduce the error rate of the features extraction of the melt pool,which provides a guarantee for selective laser melting forming process monitoring and precise control of processing parameters.
关键词
选区激光熔融/熔池图像/熔池形貌特征/超分辨率重建Key words
selective laser melting/melt pool image/morphology of melt pool/super-resolution reconstruction引用本文复制引用
出版年
2024
NETL
NSTL
万方数据