基于SERepVGG-A2的轻量化卷积神经网络在熔池状态识别中的应用

doi:10.3901/JME.2025.03.440

机械工程学报 ›› 2025, Vol. 61 ›› Issue (3): 440-448.doi: 10.3901/JME.2025.03.440

• 数字化设计与制造 • 上一篇

扫码分享

基于SERepVGG-A2的轻量化卷积神经网络在熔池状态识别中的应用

刘翰儒^1,2, 陈桂映¹, 许泽林¹, 彭世通¹, 郭嘉楠¹, 刘伟嵬³, 王奉涛¹

1. 汕头大学智能制造技术教育部重点实验室汕头 515063;
2. 中佛罗里达大学机械与航空航天工程系奥兰多 FL32816 美国;
3. 大连理工大学机械工程学院大连 116024

收稿日期:2024-02-18 修回日期:2024-08-28 发布日期:2025-03-12
作者简介:刘翰儒，男，1997年出生。主要研究方向为增材制造缺陷智能诊断识别。E-mail:Hanru.Liu@knights.ucf.edu;陈桂映，女，1998年出生，硕士研究生。主要研究方向为增材制造缺陷智能诊断识别。E-mail:22gychen@stu.edu.cn;王奉涛(通信作者)，男，1974年出生，博士，教授，博士研究生导师。主要研究方向为增材制造过程智能监测与质量控制、机电系统智能运维与健康管理。E-mail:ftwang@stu.edu.cn
基金资助:
广东省基础与应用基础研究基金(2021A1515011989、2023A1515011164)、广东省普通高校创新团队(2020KCXTD012)和国家自然科学基金(52305544、52205110)资助项目。

Application of Lightweight Convolutional Neural Network Method Based on SERepVGG-A2 in Melt Pool State Recognition

LIU Hanru^1,2, CHEN Guiying¹, XU Zelin¹, PENG Shitong¹, GUO Jianan¹, LIU Weiwei³, WANG Fengtao¹

1. Key Laboratory of Intelligent Manufacturing Technology of Ministry of Education, Shantou University, Shantou 515063;
2. Department of Mechanical and Aerospace Engineering, University of Central Florida, Orlando FL 32816 USA;
3. School of Mechanical Engineering, Dalian University of Technology, Dalian 116024

Received:2024-02-18 Revised:2024-08-28 Published:2025-03-12

摘要/Abstract

摘要： 针对传统卷积神经网络(Convolutional neural networks, CNN)应用于熔池状态识别时，浅层卷积神经网络无法有效提取熔池特征，而深层卷积神经网络又存在参数量大和冗余度高的问题，提出一种轻量化卷积神经网络(SERepVGG-A2)熔池状态识别方法。首先针对原始熔池图片存在大量无用背景区域的问题，对熔池图片进行感兴趣区域(Region of interest，ROI)提取优化了数据集；接着为了保证模型特征提取能力并同时降低模型的复杂度，提出一种结合残差结构、注意力机制和结构参数重组化的轻量化卷积神经网络模型来识别熔池状态。以Ti-10Mo钛合金定向能量沉积过程中的在线熔池图像数据验证所提方法，结果表明所提的轻量化网络模型不仅能够有效地识别熔池状态，而且比传统的CNN方法具有更高的熔池识别效率。

关键词: 熔池状态识别, 卷积神经网络, 深度学习, 注意力机制, 结构参数重组化

Abstract: A Quasi-Aiming at the problem that the traditional convolutional neural network is applied to the state recognition of the molten pool, the shallow convolutional neural network cannot effectively extract the characteristics of the molten pool, and the deep convolutional neural network has the problems of large number of parameters and high redundancy. A lightweight convolutional neural network (SERepVGG-A2) method is quantified. Firstly, in order to solve the problem of a large number of useless background areas in the original melt pool image, the region of interest (ROI) extraction of the melt pool image is performed to optimize the data set; then, in order to ensure the model feature extraction ability and reduce the complexity of the model, a lightweight convolutional neural network model combining residual structure, attention mechanism and structural parameter reorganization is proposed to identify the molten pool state. The proposed method is verified and illustrated with the online molten pool data of Ti-10Mo titanium alloy in the directed energy deposition process. The results show that the proposed lightweight network model can not only effectively identify the molten pool characteristics according to the molten pool signal, but also it has higher melting pool identification efficiency than traditional CNN methods.

Key words: melting pool state recognition, convolutional neural network, deep learning, attention mechanism, structural parameter reorganization

中图分类号:

TG156

刘翰儒, 陈桂映, 许泽林, 彭世通, 郭嘉楠, 刘伟嵬, 王奉涛. 基于SERepVGG-A2的轻量化卷积神经网络在熔池状态识别中的应用[J]. 机械工程学报, 2025, 61(3): 440-448.

LIU Hanru, CHEN Guiying, XU Zelin, PENG Shitong, GUO Jianan, LIU Weiwei, WANG Fengtao. Application of Lightweight Convolutional Neural Network Method Based on SERepVGG-A2 in Melt Pool State Recognition[J]. Journal of Mechanical Engineering, 2025, 61(3): 440-448.

参考文献

[1] BI Guijun，SUN Chennan，CHEN Huichi，et al. Microstructure and tensile properties of superalloy in100 fabricated by micro-laser aided additive manufacturing[J]. Materials & Design，2014，60:401-408.
[2] ZHAO Tong，DAHMEN M，CAI W，et al. Laser metal deposition for additive manufacturing of aa5024 and nanoparticulate tic modified aa5024 alloy composites prepared with balling milling process[J]. Optics & Laser Technology，2020，131:106438.
[3] WANG Lu，ZHANG Yaoming，CHIA H Y，et al. Mechanism of keyhole pore formation in metal additive manufacturing[J]. NPJ Computational Materials，2022，8(1):22.
[4] MENG Lingbin，MCWILLIAMS B，JAROSINSKI W，et al. Machine learning in additive manufacturing:A review[J]. JOM，2020，72(6):2363-2377.
[5] ZHANG Ziyang，LIU Zhichao，WU Dazhong. Prediction of melt pool temperature in directed energy deposition using machine learning[J]. Additive Manufacturing，2021，37:101692.
[6] MOZAFFAR M，PAUL A，AL-BAHRANI R，et al. Data-driven prediction of the high-dimensional thermal history in directed energy deposition processes via recurrent neural networks[J]. Manufacturing Letters，2018，18:35-39.
[7] KHANZADEH M，CHOWDHURY S，TSCHOPP M A，et al. In-situ monitoring of melt pool images for porosity prediction in directed energy deposition processes[J]. IISE Transactions，2019，51(5):437-455.
[8] SHEVCHIK S A，KENEL C，LEINENBACH C，et al. Acoustic emission for in situ quality monitoring in additive manufacturing using spectral convolutional neural networks[J]. Additive Manufacturing，2018，21:598-604.
[9] YE Dongsen，HSI FUH J Y，ZHANG Yingjie，et al. In situ monitoring of selective laser melting using plume and spatter signatures by deep belief networks[J]. ISA Transactions，2018，81:96-104.
[10] ZHANG Bin，LIU Shunyu，SHIN Y C. In-process monitoring of porosity during laser additive manufacturing process[J]. Additive Manufacturing，2019，28:497-505.
[11] MI Jiqian，ZHANG Yikai，LI Hui，et al. In-situ monitoring laser based directed energy deposition process with deep convolutional neural network[J]. Journal of Intelligent Manufacturing，2021.
[12] SEMAK V V，HOPKINS J A，MCCAY M H，et al. Melt pool dynamics during laser welding[J]. Journal of Physics D:Applied Physics，1995，28(12):2443-2450.
[13] GUAN Yong，MENG Zong，SUN Dengyun，et al. Rolling bearing fault diagnosis based on information fusion and parallel lightweight convolutional network[J]. Journal of Manufacturing Systems，2022，65:811-821.
[14] LÓPEZ DE LA ROSA F，GÓMEZ-SIRVENT J L，MORALES R，et al. Defect detection and classification on semiconductor wafers using two-stage geometric transformation-based data augmentation and SqueezeNet lightweight convolutional neural network[J]. Computers & Industrial Engineering，2023，183:109549.
[15] DING Xiaohan，ZHANG Xiangyu，MA Ningning，et al. RepVGG:making vgg-style convnets great again[EB/OL]. arXiv，2021.
[16] HU Jie，SHEN Li，SUN Gang. Squeeze-and-excitation networks:arXiv:1709.01507[EB/OL]. arXiv，2019.
[17] 田娟，李英祥，李彤岩. 激活函数在卷积神经网络中的对比研究[J]. 计算机系统应用，2018，27(7):43-49. TIAN Juan，LI Yingxiang，LI Tongyan. Comparative study of activation functions in convolutional neural networks [J]. Application of Computer Systems，2018，27(7):43-49.
[18] IOFFE S，SZEGEDY C. Batch normalization:Accelerating deep network training by reducing internal covariate shift[C]//International conference on machine learning. pmlr，2015:448-456.
[19] LIU Hanhu，YUAN Junlin，PENG Shitong，et al. In-suit monitoring melt pool states in direct energy deposition using resnet[J]. Measurement Science and Technology，2022，33(12):124007.
[20] OTSU N. A tlreshold selection method from gray-level histograms[J]. IEEE Transactions on Systrems，Man，and Cybernetics，1979，9(1):5.
[21] ZHOU Bolei，KHOSLA A，LAPEDRIZA A，et al. Learning deep features for discriminative localization[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas，NV，USA:IEEE，2016:2921-2929.

基于SERepVGG-A2的轻量化卷积神经网络在熔池状态识别中的应用

Application of Lightweight Convolutional Neural Network Method Based on SERepVGG-A2 in Melt Pool State Recognition

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

本文评价

[1]	黄思翰, 陈建鹏, 徐哲, 阎艳, 王国新. 基于大语言模型和机器视觉的智能制造系统人机自主协同作业方法[J]. 机械工程学报, 2025, 61(3): 130-141.
[2]	陈国强, 申正义, 杨宇驰, 李腾, 徐丽. FCN优化GA的飞行模拟座舱意象仿生视觉焦点预测模型[J]. 机械工程学报, 2025, 61(3): 154-166.
[3]	狄子钧, 袁东风, 李东阳, 梁道君, 周晓天, 信苗苗, 曹凤, 雷腾飞. 基于多尺度-高效通道注意力网络的刀具故障诊断方法[J]. 机械工程学报, 2024, 60(6): 82-90.
[4]	李春凯, 王嘉昕, 石玗, 代悦. GTAW熔池激光条纹动态行为与熔透预测方法研究[J]. 机械工程学报, 2024, 60(6): 236-244.
[5]	王志颖, 李天福, 许文纲, 孙闯, 张军辉, 徐兵, 严如强. 降噪混合注意力变分自编码器及其在轴向柱塞泵故障诊断中的应用[J]. 机械工程学报, 2024, 60(4): 167-177.
[6]	史丽晨, 史炜椿, 王海涛, 李金阳, 刘亚雄. 基于DRSN-BiLSTM模型的刀具磨损预测方法研究[J]. 机械工程学报, 2024, 60(24): 66-74.
[7]	赵德望, 姜超, 赵延广, 杨文平, 樊俊铃. 铝锂合金疲劳性能及预测研究-基于试验和“浅层”+“深度”混合神经网络方法[J]. 机械工程学报, 2024, 60(16): 190-199.
[8]	付玲, 佘玲娟, 颜镀镭, 张鹏, 龙湘云. 基于内嵌物理信息与注意力机制BiLSTM神经网络的臂架系统疲劳损伤预测模型[J]. 机械工程学报, 2024, 60(13): 205-215.
[9]	袁静, 任港星, 蒋会明, 赵倩, 魏臣隽, 朱骏. 基于多元提升核神经网络的机械故障诊断方法及其特征提取可解释性研究[J]. 机械工程学报, 2024, 60(12): 51-64.
[10]	陈钱, 陈康康, 董兴建, 皇甫一樊, 彭志科, 孟光. 一种面向机械设备故障诊断的可解释卷积神经网络[J]. 机械工程学报, 2024, 60(12): 65-76.
[11]	刘岳开, 王天杨, 褚福磊. 基于低延迟可解释性深度学习的复杂旋转机械关键部件知识嵌入与诊断方法研究[J]. 机械工程学报, 2024, 60(12): 107-115.
[12]	王振亚, 刘韬, 伍星. 互信息规范的卷积神经网络及其在轴承故障特征提取中的应用[J]. 机械工程学报, 2024, 60(12): 137-146.
[13]	赖旭伟, 丁昆, 张楷, 黄锋飞, 郑庆, 李致萱, 丁国富. 基于可解释物理引导空间注意力改进的跨工艺参数立铣刀磨损辨识[J]. 机械工程学报, 2024, 60(12): 147-157.
[14]	林昕, 毛杨坤, 傅盈西, 朱锟鹏. 基于图像语义分割与图卷积的选区激光熔融成形过程羽流运动特征分析[J]. 机械工程学报, 2024, 60(12): 168-182.
[15]	邵海东, 肖一鸣, 邓乾旺, 任颖莹, 韩特. 基于不确定性感知网络的可信机械故障诊断[J]. 机械工程学报, 2024, 60(12): 194-206.