基于视觉传感的热丝激光金属沉积熔滴—熔池多特征信息同步监测

李春凯; 潘宇; 石玗; 王文楷; 赵中博

doi:10.12073/j.hjxb.20240711002

基于视觉传感的热丝激光金属沉积熔滴—熔池多特征信息同步监测

李春凯^1,,
潘宇¹,
石玗^{1, 2},
王文楷¹,
赵中博¹

1.
兰州理工大学, 省部共建有色金属先进加工与再利用国家重点实验室, 兰州 , 730050
2.
兰州理工大学, 有色金属合金及加工教育部重点实验室, 兰州 , 730050

基金项目: 国家自然科学基金资助项目(52365048)；甘肃省高校科研创新平台重大培育项目(2024CXPT-06)；甘肃省科技重大专项(22ZD6GA008)

详细信息

作者简介:
李春凯，博士，副研究员；主要研究方向为智能焊接及焊接过程自动化；Email：15339316249@163.com

中图分类号: TG 456.7；TP 273
计量
- 文章访问数: 154
- HTML全文浏览量: 9
- PDF下载量: 55
出版历程
- 收稿日期: 2024-07-10
- 网络出版日期: 2024-10-23
- 刊出日期: 2024-11-24

Feature information extraction of hot wire laser metal deposition process based on object and key point detection

1.
State Key Laboratory of Advanced Processing and Recycling of Non-ferrous Metals, Lanzhou University of Technology, Lanzhou, 730050, China
2.
Key Laboratory of Non-ferrous Metal Alloys and Processing of State Education Ministry, Lanzhou University of Technology, Lanzhou, 730050, China

摘要

摘要:
为了提高热丝激光金属沉积(HW-LMD)过程中的质量稳定性和实现熔滴—熔池多特征信息的同步实时监测，采用基于高动态视觉相机结合YOLO v8深度学习神经网络的高精度实时监控方法，通过相机捕捉HW-LMD过程中的动态变化，并利用YOLO v8神经网络对过渡方式和熔池行为进行同步监测，首先判断沉积过程是否为稳定的液桥过渡，然后在液桥过渡模式下提取熔池尺寸的关键点信息. 结果表明，YOLO v8神经网络在检测沉积过程过渡方式和熔池关键点信息方面具有高精确度，精确率分别达到了98.8%和99.9%，熔池宽度的平均误差为4.1%，且推理时间平均仅为12 ms/帧，满足了HW-LMD过程实时监控的需求.
- 热丝激光金属沉积 /
- 过程监控 /
- YOLO v8深度学习神经网络 /
- 熔池尺寸 /
- 过渡方式
Abstract:
In order to improve the quality stability and achieve simultaneous real-time monitoring of droplet-melting pool multi-feature information during the hot-wire laser metal deposition (HW-LMD) process, this study adopts a high-precision real-time monitoring method based on a high-dynamic vision camera combined with a YOLO v8 deep learning neural network. The dynamic changes in the HW-LMD process are captured by a camera, and the transition mode and melt pool behaviour are monitored synchronously using the YOLO v8 neural network, which firstly determines whether the deposition process is a stable liquid bridge transition or not, and then extracts the key point information of the melt pool dimensions in the liquid bridge transition mode. The results show that the YOLO v8 neural network has high accuracy in detecting the transition mode and melt pool key point information of the deposition process, with an accuracy rate of 98.8% and 99.9%, respectively, and an average error of 4.1% for the melt pool width, and the inference time is only 12 ms per frame on average, which meets the demand for real-time monitoring of the HW-LMD process.
- hot wire laser metal deposition /
- process monitoring /
- YOLO v8 deep learning neural network /
- melt pool size /
- transition mode

HTML全文

图 1 HW-LMD试验系统

Figure 1. HW-LMD experimental system

下载: 全尺寸图片幻灯片

图 2 不同过渡方式图像

Figure 2. Transition modes images. (a) droplet transition; (b) liquid bridge transition

下载: 全尺寸图片幻灯片

图 3 数据集标注

Figure 3. Dataset annotation. (a) droplet transition; (b) liquid bridge transition

下载: 全尺寸图片幻灯片

图 4 过程特征信息提取流程

Figure 4. Process of feature information extraction

下载: 全尺寸图片幻灯片

图 5 校准平面和熔池尺寸计算示意图

Figure 5. Calibration planes and schematic diagram of melt pool size calculation. (a) calibration planes; (b) schematic diagram of melt pool size

下载: 全尺寸图片幻灯片

图 6 训练结果

Figure 6. Training results. (a) training Loss; (b) validation Loss; (c) object detection; (d) key point detection

下载: 全尺寸图片幻灯片

图 7 预测结果

Figure 7. Predicted results. (a) confusion matrix; (b) melt pool width prediction error

下载: 全尺寸图片幻灯片

图 8 模型推理结果

Figure 8. Model inference results. (a) liquid bridge transition; (b) droplet transition

下载: 全尺寸图片幻灯片

参考文献(12)

[1]	Abuabiah M, Mbodj N G, Shaqour B, et al. Advancements in laser wire-feed metal additive manufacturing: A brief review[J]. Materials, 2023, 16(5): 2030 − 2053.
[2]	产玉飞, 陈长军, 张敏. 金属增材制造过程的在线监测研究综述[J]. 材料导报, 2019, 33(17): 2839 − 2846. doi: 10.11896/cldb.19010111 Chan Yufei, Chen Changjun, Zhang Min. Review of on-line monitoring research on metal additive manufacturing process[J]. Materials Review, 2019, 33(17): 2839 − 2846. doi: 10.11896/cldb.19010111
[3]	Zhao Y, Yang Q, Huang J, et al. Droplet transfer and weld geometry in laser welding with filling wire[J]. The International Journal of Advanced Manufacturing Technology, 2017, 90(5-8): 2153 − 2161. doi: 10.1007/s00170-016-9486-3
[4]	郑世卿, 温鹏, 单际国. 激光热丝焊接过程焊丝过渡行为及其稳定性的研究[J]. 中国激光, 2014, 41(4): 109 − 116. Zheng Shiqing, Wen Peng, Shan Jiguo. Research on wire transfer and its stability in laser hot wire welding process[J]. Chinese Journal of Lasers, 2014, 41(4): 109 − 116.
[5]	Chang S, Gach S, Senger A, et al. A new 3D printing method based on non-vacuum electron beam technology[J]. Journal of Physics: Conference Series, 2018, 1074(1): 012017.
[6]	Chang S, Zhang H, Xu H, et al. Closed-loop control of droplet transfer in electron-beam free form fabrication[J]. Sensors, 2020, 20(3): 923 − 935.
[7]	Zhang Z, Kong F, Kovacevic R. Laser hot-wire cladding of Co-Cr-W metal cored wire[J]. Optics and Lasers in Engineering, 2020, 128: 105998 − 106010.
[8]	Kisielewicz A, Mi Y, Sikström F, et al. Multi sensor monitoring of the wire-melt pool interaction in hot-wire directed energy deposition using laser beam[J]. IOP Conference Series: Materials Science and Engineering, 2023, 1296(1): 012011. doi: 10.1088/1757-899X/1296/1/012011
[9]	郭忠峰, 刘俊池, 杨钧麟. 基于关键点检测方法的焊缝识别[J]. 焊接学报, 2024, 45(1): 88 − 93. doi: 10.12073/j.hjxb.20230204001 Guo Zhongfeng, Liu Junchi, Yang Junlin. Weld recognition based on key point detection method[J]. Transactions of the China Welding Institution, 2024, 45(1): 88 − 93. doi: 10.12073/j.hjxb.20230204001
[10]	Chandra M, Rajak S, Vimal K E K. Deep learning-based framework for the observation of real-time melt pool and detection of anomaly in wire-arc additive manufacturing[J]. Materials and Manufacturing Processes, 2024, 39(6): 761 − 777. doi: 10.1080/10426914.2023.2254386
[11]	Jocher G, Chaurasia A, Qiu J. YOLO by Ultralytics[Z]. Frederick: Ultralytics, 2023.
[12]	Wang W, Shi Y, Li C, et al. Feature information extraction method for narrow gap U-type groove based on laser vision[J]. Journal of Manufacturing Processes, 2023, 104: 257 − 272. doi: 10.1016/j.jmapro.2023.08.053

施引文献(8)

期刊类型引用(3)

1.	李麒，朱光平. 基于SWT和SVR的重力坝变形预测研究. 人民长江. 2021(11): 169-174 . 百度学术
2.	左曙光，潘健，吴旭东，冯朝阳. 考虑动圈偏心的电动振动台等效电磁力计算方法. 西安交通大学学报. 2020(08): 132-139 . 百度学术
3.	叶建雄，彭星玲，李兵. 水下湿法焊接研究进展. 电焊机. 2020(09): 111-117 . 百度学术