基于深度学习的激光熔覆熔池形貌提取方法研究
Research on Extraction Method of Molten Pool Morphology in Laser Cladding Based on Deep Learning
雍耀维 1张伟伟 1王刚 1张帅 1刘海波2
作者信息
- 1. 宁夏大学机械工程学院,宁夏银川 750021
- 2. 吴忠仪表有限责任公司,宁夏吴忠 751100
- 折叠
摘要
激光熔覆技术是先进制造的重要发展方向.随着科技和工业的高速发展,人们对熔覆层的质量要求越来越高,熔池作为激光熔覆过程中一个重要的检测对象,其特征直接决定着熔覆层的好坏.目前熔池形貌的提取都是基于传统的图像分割实现的,遇到特征不明显的熔池时,传统图像分割表现不佳,影响后续实验结果.为此,开发一种基于语义分割网络的熔池形貌提取方法,用语义分割网络替代传统图像分割,结合机器语言实现熔池形貌的高精度提取,并用实验验证证明方法的可行性,为后续熔池的检测与分析打下坚实基础.
Abstract
Laser cladding technology is an promising development direction of advanced manufacturing.With the rapid develop-ment of technology and industry,the quality of the cladding layer is becoming more and more demanding.The characteristics of the melt pool,as an important test object in the laser cladding process,directly determine the quality of the cladding layer.Currently the extraction of melt pool profiles is based on traditional image segmentation,which does not perform well when en-countering melt pools with obscure features.Therefore,this work developed a molten pool morphology extraction method based on semantic segmentation network.The semantic segmentation network is used instead of traditional image segmenta-tion,combined with machine language to achieve high-precision extraction of melt pool shape and experimental verification to prove the feasibility of the method,laying a solid foundation for the subsequent detection and analysis of the melt pool.
关键词
深度学习/语义分割/激光熔覆/熔池检测/图像处理Key words
deep learning/semantic segmentation/laser cladding/molten pool detection/image processing引用本文复制引用
基金项目
宁夏回族自治区重点研发计划(引才专项)(2020BEB04030)
出版年
2024
NETL
NSTL
万方数据