基于YOLO算法的激光粉末床熔融成形层形貌分类识别与成形质量预测研究

doi:10.3901/JME.260081

摘要/Abstract

摘要： 激光粉末床熔融(Laser powder bed fusion, LPBF)技术能够高精度制造复杂金属构件，其成形过程的质量波动与缺陷在线监测是目前研究的重点方向之一。本研究面向Ti-6Al-4V合金LPBF过程，构建了一种基于原位视觉感知的成形层形貌在线监测与分类识别方法，可实现对成形质量的预测。首先，通过单道熔道实验系统分析不同激光功率与扫描速度组合下的熔池行为及成形层光学形貌特征，将成形层形貌依据能量密度划分为低能区、适能区与高能区，为后续分类标注建立实验基准。随后开展9组不同工艺参数的成形实验，并采集逐层成形图像，表征成形质量，构建“工艺参数—成形层形貌—成形质量”之间的定量关联。基于采集的图像数据构建多模态增强数据集(包括几何增强、噪声注入与光照调整)，并采用YOLOv5s模型学习成形层光学特征与能量输入状态之间的映射关系，实现对成形质量区间的在线识别与预测。实验结果表明，模型在100个Epoch训练后，可对高、中、低能量密度形貌的识别达到 97% 以上准确率(mAP>0.90)。研究揭示了成形工艺参数驱动下的成形质量与成形层光学形貌之间的对应关系，为LPBF过程质量在线监测与实时调控提供了可工程化的技术路径。

关键词: 激光粉末床熔融, 在线监测, 深度学习, 成形层图像分类识别, YOLO算法, 成形质量预测

Abstract: Laser powder bed fusion (LPBF) enables the high-precision fabrication of complex metal components; quality fluctuations during the process and the lack of reliable in-situ defect monitoring is one of the key research directions. Focusing on the LPBF fabrication of Ti-6Al-4V alloys, this study develops an in-situ vision-based method for online monitoring and classification of layer-wise build quality, which enables the prediction of forming quality. First, single-track experiments are conducted to investigate melt-pool behavior and optical morphology under different combinations of laser power and scanning velocity. Based on the resulting energy-density variations, the layer morphology is categorized into low-energy, optimal-energy, and high-energy regimes, establishing ground-truth criteria for subsequent classification. Nine groups of multi-parameter LPBF experiments are then performed to acquire layer-wise images and characterize the corresponding build quality, enabling the construction of a quantitative relationship among processing parameters, surface morphology, and final part quality. A multimodal augmented dataset (including geometric transformations, noise injection, and illumination adjustment) is built using the collected images. The YOLOv5s network is employed to learn the mapping between optical features of the layer morphology and the energy-input state, achieving real-time online recognition and prediction of quality categories. Experimental results show that after 100 training epochs, the model achieves over 97% classification accuracy (mAP > 0.90) for distinguishing high, medium, and low energy-density morphologies. This work elucidates the correspondence between LPBF processing parameters, optical layer morphology, and resultant build quality, providing an engineering-feasible pathway for online monitoring and real-time control of LPBF processes.

Key words: laser powder bed fusion (LPBF), in-situ monitoring, deep learning, layer-wise image classification, YOLO algorithm, prediction of forming quality

中图分类号:

TG665

李雨露, 李俊峰, 万章艺, 魏正英. 基于YOLO算法的激光粉末床熔融成形层形貌分类识别与成形质量预测研究[J]. 机械工程学报, 2026, 62(3): 218-234.

LI Yulu, LI Junfeng, WAN Zhangyi, WEI Zhengying. Study on Classification and Recognition of Laser Powder Bed Fusion Forming Layer Morphology and Prediction of Forming Quality Based on YOLO Algorithm[J]. Journal of Mechanical Engineering, 2026, 62(3): 218-234.

导出引用管理器 EndNote|Reference Manager|ProCite|BibTeX|RefWorks

链接本文: http://www.cjmenet.com.cn/CN/10.3901/JME.260081

http://www.cjmenet.com.cn/CN/Y2026/V62/I3/218

参考文献

[1] 赵德陈，林峰．金属粉末床熔融工艺在线监测技术综述[J]．中国机械工程，2018，29(17)：2100-2110，2118． ZHAO Dechen，LIN Feng．A review of on line monitoring techniques in metal powder bed fusion processes[J]．China Mechanical Engineering，2018，29(17)：2100-2110，2118．
[2] 卢秉恒. 增材制造技术—现状与未来[J]. 中国机械工程，2020，31(1)：19-23. LU Bingheng. Additive manufacturing：Current situation and future[J]. China Mechanical Engineering，2020，31(1)：19-23.
[3] ZHU Sheng，DU Wenbo，WANG Xiaoming，et al. Advanced additive remanufacturing technology[J]. Chinese Journal of Mechanical Engineering: Additive Manufacturing Frontiers，2023，2(1)：100066.
[4] WANG J，ZHU R，LIU Y，et al. Understanding melt pool characteristics in laser powder bed fusion：An overview of single-and multi-track melt pools for process optimization[J]. Advanced Powder Materials，2023，2(4)：100137.
[5] 李瑞迪，魏青松，刘锦辉，等. 选择性激光熔化成形关键基础问题的研究进展[J]. 航空制造技术，2012(5)：26-31. LI Ruidi，WEI Qingsong，LIU Jinhui，et al. Research progress on key fundamental issues of selective laser melting forming[J]. Aeronautical Manufacturing Technology，2012(5)：26-31.
[6] 赵冬华，张国权，阮凯程，等. 面向航空航天轻量化结构的机器人化增材制造技术研究进展[J]. 航空制造技术，2024，67(19)：71-91. ZHAO Donghua，ZHANG Guoquan，RUAN Kaicheng，et al. Research progress on robotic additive manufacturing technology for aerospace lightweight structures[J]. Aeronautical Manufacturing Technology，2024，67(19)：71-91.
[7] 宋波，张磊，王晓波，等. 面向航空航天的增材制造超材料的研究现状及发展趋势[J]. 航空制造技术，2022，65(14)：22-33. SONG Bo，ZHANG Lei，WANG Xiaobo，et al. Research status and development trend of additive manufacturing metamaterials for aerospace[J]. Aeronautical Manufacturing Technology，2022，65(14)：22-33.
[8] 杨胶溪，吴文亮，王长亮，等. 激光选区熔化技术在航空航天领域的发展现状及典型应用[J]. 航空材料学报，2021，41(2)：1-15. YANG Jiaoxi，WU Wenliang，WANG Changliang，et al. Development status and typical applications of selective laser melting technology in aerospace field[J]. Journal of Aeronautical Materials，2021，41(2)：1-15.
[9] CHEN Y，CLARK S J，LEUNG C L A，et al. In-situ synchrotron imaging of keyhole mode multi-layer laser powder bed fusion additive manufacturing[J]. Applied Materials Today，2020，20：100650.
[10] WILLIAMS R J，PIGLIONE A，R O NNEBERG T，et al. In situ thermography for laser powder bed fusion：Effects of layer temperature on porosity，microstructure and mechanical properties[J]. Additive Manufacturing，2019，30：100880.
[11] 赵志斌，王晨希，张兴武，等. 激光粉末床熔融增材制造过程智能监控研究进展与挑战[J]. 机械工程学报，2023，59(19)：253-276. ZHAO Zhibin，WANG Chenxi，ZHANG Xingwu ，et al. Research progress and challenges in process intelligent monitoring of laser powder bed fusion additive manufacturing[J]. Journal of Mechanical Engineering，2023，59(19)：253-276.
[12] ZHANG B，ZIEGERT J，FARAHI F，et al. In situ surface topography of laser powder bed fusion using fringe projection[J]. Additive Manufacturing，2016，12：100-107.
[13] CALTANISSETTA F，GRASSO M，PETRÒ S，et al. Characterization of in-situ measurements based on layerwise imaging in laser powder bed fusion[J]. Additive Manufacturing，2018，24：183-199.
[14] HE P，ZHONG K，LIU X，et al. A phase-guided method for extracting the contour of the fusion area in laser powder bed fusion[C].//Seventh International Conference on Optical and Photonic Engineering (icOPEN 2019). SPIE，2019，11205：307-312.
[15] KORZENIOWSKI M，MAŁACHOWSKA A M，SZYMAŃSKI M. Development and implementation of a deep learning algorithm to evaluate the powder distribution process during 3D printing using the LPBF method[J]. Applied Sciences. 2024，14(24)：11718.
[16] WEI J，HE Y，WANG F，et al. Convolutional neural network assisted infrared imaging technology：an enhanced online processing state monitoring method for laser powder bed fusion[J]. Infrared Physics & Technology，2023，131：104661.
[17] 吴世彪，窦文豪，杨永强，等. 面向激光选区熔化金属增材制造的检测技术研究进展[J]. 精密成形工程，2019，11(4)：37-50. WU Shibiao，DOU Wenhao，YANG Yongqiang，et al. Research progress on detection technology for selective laser melting metal additive manufacturing[J]. Journal of Netshape Forming Engineering，2019，11(4)：37-50.
[18] CHEN L，BI G，YAO X，et al. In-situ process monitoring and adaptive quality enhancement in laser additive manufacturing：A critical review[J]. Journal of Manufacturing Systems，2024，527-74.
[19] JIAO Mohan，LONG Hongyu，XIAO Buwei，et al. Electron beam powder bed fusion additive manufacturing：A comprehensive review and its development in China[J]. Additive Manufacturing Frontiers，2024，3(4)：200177.
[20] 徐晓静. 基于机器视觉的激光选区熔化铺粉缺陷自动检测与分类研究[D]. 武汉：华中科技大学，2021. XU Xiaojing. Research on automatic detection and classification of powder spreading defects in selective laser melting based on machine vision[D]. Wuhan：Huazhong University of Science and Technology，2021.