| 引用本文: | 贾琪,何建萍,李芳,华学明.基于机器学习的薄板等离子弧搭接焊的熔深预测[J].材料科学与工艺,2023,31(4):51-59.DOI:10.11951/j.issn.1005-0299.20220247. |
| JIA Qi,HE Jianping,LI Fang,HUA Xueming.Penetration depth predicition of thin plate plasma arc lap welding based on machine learning[J].Materials Science and Technology,2023,31(4):51-59.DOI:10.11951/j.issn.1005-0299.20220247. |
|
| 摘要: |
| 在等离子弧搭接焊中,搭接焊接头的焊缝熔深是评价焊接质量的关键指标之一,而焊接过程中的热输入信息和熔池图像信息都与焊缝熔深有密切关系。本文通过建立304L不锈钢薄板等离子弧搭接焊数据采集系统,利用LabVIEW实时检测电信息,采用视觉传感技术实时获取薄板等离子弧搭接焊过程中的熔池图像,并通过图像处理方法获得熔池的几何参数信息,结合焊接工艺参数,选择峰值电流、峰值电压、焊接速度、离子气流量、保护气流量、熔池宽度和熔池后端长度作为输入量,焊缝熔深作为输出量,建立了基于支持向量机回归和BP神经网络的熔深预测模型。实验验证表明,采用径向基函数的支持向量机回归模型可以有效地对焊缝熔深进行预测,并具有很好的泛化能力,可为进一步实现在线优化焊接工艺参数提供依据。 |
| 关键词: 等离子弧焊 薄板 LabVIEW 视觉传感 SVR 熔深预测 |
| DOI：10.11951/j.issn.1005-0299.20220247 |
| 分类号:TG456.2;TP181 |
| 文献标识码:A |
| 基金项目:国家自然科学基金资助项目(51775327)；工业和信息化部高技术船舶科研计划2020函313号《MARK III型液货围护系统建造工艺与关键技术研究》. |
|
| Penetration depth predicition of thin plate plasma arc lap welding based on machine learning |
|
JIA Qi1,HE Jianping1,LI Fang2,HUA Xueming2
|
|
(1.College of Materials Science and Engineering, Shanghai University of Engineering Science, Shanghai 201620,China;2.Shanghai Key Labs of Laser Manufacturing and Material Modification(Shanghai Jiao Tong University), Shanghai 200240,China)
|
| Abstract: |
| In plasma arc lap welding, the weld depth of the lap weld head is one of the key indicators to evaluate the welding quality,and both the heat input information and the melt pool image information during the welding process are closely related to the weld depth. In this paper, the data acquisition system of 304L stainless steel sheet plasma arc lap welding was established; the electrical information was detected in real-time by LabVIEW; the visual sensing technology was used to obtain real-time images of the melt pool in the thin plate plasma arc lap welding process, and the geometric parameters information of the melt pool was abtained through image processing methods.A melt depth prediction model based on support vector machine regression and BP neural network was established, combined with the welding process parameters, the selection of peak current, peak voltage, welding speed, ion gas flow, shielding gas flow rate, melt pool width and melt pool back end length as the input quantity, weld depth as the output quantity. Experimental verification shows that the support vector machine regression model using radial basis functions can effectively predict the weld depth of melt and has a good generalization ability, which can provide a basis for further online optimization of welding process parameters. |
| Key words: plasma arc welding thin plate LabVIEW visual sensing SVR penetration prediction |
