融合空间和通道特征的等离子弧焊熔池熔透状态预测方法

陈宸; 周方正; 李成龙; 刘新锋; 贾传宝; 徐瑶

doi:10.12073/j.hjxb.20220516001

融合空间和通道特征的等离子弧焊熔池熔透状态预测方法

1.
山东建筑大学, 济南, 250101
2.
山东大学, 济南, 250100

基金项目: 国家自然科学基金资助项目(51975332,62102235)；山东建筑大学博士基金资助项目(X19023Z0101).

详细信息

作者简介:
陈宸，硕士；主要从事工业大数据研究； Email: 819221454@qq.com

通讯作者:
刘新锋，博士，副教授；Email: liuxinfeng18@sdjzu.edu.cn

中图分类号: TG 444
计量
- 文章访问数: 277
- HTML全文浏览量: 31
- PDF下载量: 52
出版历程
- 收稿日期: 2022-05-15
- 网络出版日期: 2023-04-05
- 刊出日期: 2023-04-24

Prediction method of plasma arc welding molten pool melting state based on spatial and channel characteristics

1.
Shandong Jianzhu University, Jinan, 250101, China
2.
Shandong University, Jinan, 250100, China

摘要

摘要: 为了提高等离子弧焊熔池熔透状态预测的准确率，满足工业应用的需求，提出了一种融合图像空间和通道特征的熔池熔透状态预测模型PCSCNet. 在该模型中对残差网络(residual network, ResNet50)结构进行改造，并融入压缩和激励网络来同时提取熔池正面图像的空间和通道特征信息. 采用恒定电流等离子弧焊试验的数据集进行测试，建立了熔池正面图像与熔池熔透状态的对应关系. 结果表明，模型预测准确率提升到95%以上. 采用Grad-CAM方法对模型进行可视化，分析并揭示了模型预测的聚焦区域，与实际熔池的图像特征进行对比，验证了模型的合理性.
- 深度学习 /
- 等离子弧焊 /
- 空间和通道特征 /
- 压缩和激励网络 /
- 残差网络
Abstract: To improve the accuracy of predicting the molten pool penetration state in plasma arc welding so as to meet industrial needs, this paper proposes a model called PCSCNet that integrates image space and channel characteristics. In this model, the convolutional residual network ResNet50 structure is modified and integrated into the channel attention network squeeze and excitation network to simultaneously extract spatial feature information and channel feature information from the front image of the molten pool. By testing on a dataset of constant current plasma arc welding experiments, the model establishes the corresponding relationship between the front surface image of the weld pool and the state of the keyhole. The results show that the model achieves a prediction accuracy of over 95%. Using the Grad-CAM method, the model's predicted focus area is visualized, analyzed, and compared with the actual molten pool's image features to verify the model's reliability.
- deep learning /
- plasma arc welding /
- spacial and channel features /
- squeeze and excitation network /
- residual network

HTML全文

图 1 等离子弧视觉检测平台示意图

Figure 1. Diagram of plasma arc visual detection platform

下载: 全尺寸图片幻灯片

图 2 数据增强图样例

Figure 2. Sample diagram of data enhancement

下载: 全尺寸图片幻灯片

图 3 CNN结构示意图

Figure 3. CNN structure diagram

下载: 全尺寸图片幻灯片

图 4 残差块单元结构

Figure 4. Residual block element structure

下载: 全尺寸图片幻灯片

图 5 SE模块的总体结构

Figure 5. Overall structure of SE module

下载: 全尺寸图片幻灯片

图 6 ResNet50模型改进

Figure 6. ResNet50 model improvement

下载: 全尺寸图片幻灯片

图 7 Se-block 结构

Figure 7. Se-block structure

下载: 全尺寸图片幻灯片

图 8 Se-block嵌入位置

Figure 8. Se-block embedding location

下载: 全尺寸图片幻灯片

图 9 模型训练曲线

Figure 9. Model training curve. (a) loss of PCSCNet training and validation; (b) accuracy of PCSCNet training and validation

下载: 全尺寸图片幻灯片

图 10 正面熔池图像特征

Figure 10. Front molten pool image feature

下载: 全尺寸图片幻灯片

图 11 Grad-CAM可视化结果

Figure 11. Grad-cam visualization results. (a) original diagram; (b) thermodynamic diagram; (c) Grad-CAM diagram

下载: 全尺寸图片幻灯片

表 1 焊接工艺参数

Table 1 Welding paraments

试验编号	氩气流量 Q/(L·min⁻¹)	焊接电流 I/A	焊接速度 v/(mm·min⁻¹)	电弧电压 U/ V
1	2.6	125	120	29.12
2	2.6	130	120	30.06
3	2.6	135	120	30.99
4	2.6	140	120	31.93
5	2.6	145	120	32.86
6	2.6	150	120	33.80
7	2.6	125	90	35.10
8	2.6	125	100	35.24
9	2.6	125	110	35.38
10	2.6	125	120	35.52
11	2.6	125	130	35.66
12	2.4	125	120	29.79
13	2.6	125	120	30.02
14	2.8	125	120	30.24
15	3.0	125	120	30.47
16	3.2	125	120	30.69

下载: 导出CSV

表 2 数据集中样本数量

Table 2 Sample quantity in data set 个

数据集	穿孔状态样本数n₀	未穿孔状态样本数n₁
训练集	3 267	2 801
验证集	1 088	946
测试集	1 088	946

下载: 导出CSV

表 3 试验参数设置

Table 3 Experimental parameter setting

试验编号	优化器	学习率 l_r	批大小 p	迭代次数 N/次	精度 P(%)
1	SGD	0.001	32	100	59.14
2	Adam	0.001	32	100	95.47
3	Adam	0.01	32	100	53.56
4	Adam	0.000 1	32	100	95.84
5	Adam	0.000 1	64	100	95.57
6	Adam	0.000 1	16	100	95.28
7	Adam	0.000 1	32	200	95.17
8	Adam	0.000 1	32	50	95.15

下载: 导出CSV

表 4 不同网络结构对预测性能的影响

Table 4 Effects of different network structures on prediction performance 　　　　　　　 %

模型名称	准确率A	精度P	召回率R	F₁
ResNet50	93.02	93.24	89.17	91.16
SE-ResNet50	94.64	94.81	94.71	94.76
PCSCNet	95.55	93.42	96.19	94.79

下载: 导出CSV

表 5 模型在测试集上的性能表现

Table 5 Performance of the model on the test set 　 %

模型名称	准确率A	精度P	召回率R	F₁
VGG16	90.28	88.50	92.80	90.60
ResNet50	93.02	93.24	89.17	91.16
InceptionNetV3	94.59	91.28	97.16	94.27
PCSCNet	95.55	93.42	96.19	94.79
AlexNet-trans	94.60	80.00	94.10	86.48

下载: 导出CSV

参考文献(20)

[1]	陈健, 苏金花, 张毅梅. 《中国制造2025》与先进焊接工艺及装备发展[J]. 焊接, 2016(3): 1 − 5. doi: 10.3969/j.issn.1001-1382.2016.03.001 Chen Jian, Su Jinhua, Zhang Yimei. 《Made in China 2025》and the development of advanced welding technology and equipment[J]. Welding & Joining, 2016(3): 1 − 5. doi: 10.3969/j.issn.1001-1382.2016.03.001
[2]	忻建文, 吴东升, 李芳, 等. 小孔型等离子弧焊条形气孔形成机理[J]. 焊接学报, 2021, 42(12): 54 − 61,86. doi: 10.12073/j.hjxb.20210414003 Xin Jianwen, Wu Dongsheng, Li Fang, et al. Formation mechanism of strip porosity in small hole plasma arc welding[J]. Transactions of the China Welding Institution, 2021, 42(12): 54 − 61,86. doi: 10.12073/j.hjxb.20210414003
[3]	武传松, 赵晨昱, 贾传宝. 穿孔等离子弧焊接工艺研究进展[J]. 航空制造技术, 2015, 58(20): 34 − 39. Wu Chuansong, Zhao Chenyu, Jia Chuanbao. Research progress of piercing plasma arc welding process[J]. Aeronautical Manufacturing Technology, 2015, 58(20): 34 − 39.
[4]	何建萍, 吴鑫, 吉永丰, 等. 100 μm超薄不锈钢板脉冲微束等离子弧焊成形机理[J]. 焊接学报, 2021, 42(6): 77 − 84. He Jianping, Wu Xin, Ji Yongfeng, et al. 100 μm forming mechanism of pulsed micro beam plasma arc welding of ultra-thin stainless steel plate[J]. Transactions of the China Welding Institution, 2021, 42(6): 77 − 84.
[5]	李挺, 黄健康, 陈秀娟, 等. 旁路耦合微束等离子弧热特性及焊缝成形特点[J]. 焊接学报, 2018, 39(9): 55 − 60. Li Ting, Huang Jiankang, Chen Xiujuan, et al. Thermal characteristics and weld forming characteristics of bypass coupled micro beam plasma arc[J]. Transactions of the China Welding Institution, 2018, 39(9): 55 − 60.
[6]	刘建伟, 刘媛, 罗雄麟. 深度学习研究进展[J]. 计算机应用研究, 2014, 31(7): 1921 − 1942. doi: 10.3969/j.issn.1001-3695.2014.07.001 Liu Jianwei, Liu Yuan, Luo Xionglin. Research progress of deep learning[J]. Application Research of Computers, 2014, 31(7): 1921 − 1942. doi: 10.3969/j.issn.1001-3695.2014.07.001
[7]	覃科, 刘晓刚, 丁立新. 基于卷积神经网络的CO₂焊接熔池图像状态识别方法[J]. 焊接, 2017(6): 21 − 26. Tan Ke, Liu Xiaogang, Ding Lixin. State recognition method of CO₂ welding pool image based on convolution neural network[J]. Welding & Joining, 2017(6): 21 − 26.
[8]	李海超, 刘景风, 谢吉兵, 等. 基于卷积神经网络的GTAW熔透预测[J]. 机械工程学报, 2019, 55(17): 22 − 28. doi: 10.3901/JME.2019.17.022 Li Haichao, Liu Jingfeng, Xie Jibing et al. GTAW penetration prediction based on convolutional neural networks[J]. Journal of Mechanical Engineering, 2019, 55(17): 22 − 28. doi: 10.3901/JME.2019.17.022
[9]	Li C, Wang Q, Jiao W, et al. Deep learning-based detection of penetration from weld pool reflection images[J]. Welding Journal, 2020, 99(9): 239s − 245s. doi: 10.29391/2020.99.022
[10]	刘新锋. 基于正面熔池图像和深度学习算法的PAW穿孔/熔透状态预测[D]. 济南: 山东大学, 2017. Liu Xinfeng. Prediction of PAW perforation/penetration state based on frontal molten pool image and depth learning algorithm[D]. Jinan: Shandong University, 2017.
[11]	Gu J, Wang Z, Kuen J, et al. Recent advances in convolutional neural networks[J]. Pattern Recognition, 2015, doi: 10.48550/arXiv.1512.07108.
[12]	He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA, 2016: 770 − 778.
[13]	Hu J, Shen L, Sun G. Squeeze-and-excitation networks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA, 2018: 7132 − 7141.
[14]	He T, Zhang Z, Zhang H, et al. Bag of tricks for image classification with convolutional neural networks[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Los Angeles, USA, 2019: 558 − 567.
[15]	Han S, Jeff P, John T, et al. Learning both weights and connections for efficient neural networks[C]//NIPS'15: 28th International Conference on Neural Information Processing Systems. MIT Press Cambridge,USA,2015: 1135 − 1143.
[16]	Pham H, Guan MY, Zoph B, et al. Efficient neural architecture search via parameters sharing[C]//International Conference on Machine Learning. PMLR, Stockholm, Sweden, 2018: 4095 − 4104.
[17]	吕国豪, 罗四维, 黄雅平, 等. 基于卷积神经网络的正则化方法[J]. 计算机研究与发展, 2014, 51(9): 1891 − 1900. doi: 10.7544/issn1000-1239.2014.20140266 Lyu Guohao, Luo Siwei, Huang Yaquan, et al. Regularization method based on convolution neural network[J]. Journal of Computer Research and Development, 2014, 51(9): 1891 − 1900. doi: 10.7544/issn1000-1239.2014.20140266
[18]	Szegedy C, Vanhoucke V, Ioffe S, et al. Rethinking the inception architecture for computer vision[C]//IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA, 2016: 2818 − 2826.
[19]	Jia C B, Liu X F, Zhang G K, et al. Penetration/keyhole status prediction and model visualization based on deep learning algorithm in plasma arc welding[J]. The International Journal of Advanced Manufacturing Technology, 2021, 117(11): 3577 − 3597.
[20]	Selvaraju R R, Cogswell M, Das A, et al. Grad-CAM: Visual explanations from deep networks via gradient-based localization[C]//IEEE International Conference on Computer Vision and Pattem Recognition. Hawaii, USA, 2017: 618 − 626.

施引文献(5)

期刊类型引用(3)

1.	胡继涛，马晓锋，赵荣丽，刘海生，王中任. 基于轻量级DeepLabV3+网络的焊接熔池图像分割方法. 计算机集成制造系统. 2025(01): 126-134 . 百度学术
2.	蒋凡，房世童，张国凯，陈树君，李天明，徐斌. VPPAW背面匙孔状态正面检测技术. 焊接学报. 2025(01): 8-14 . 本站查看
3.	王颖，高胜. 基于改进CeiT的GTAW焊接熔透状态识别方法. 焊接学报. 2024(04): 26-35+42+130-131 . 本站查看