CC BY
0For citation: Ren Zhenghui. Research on Weld Pool egmentation method for MAG Welding based on lightweight convolution Neural Network // Grand Altai Research & Education — Issue 2 (22)'2024 (DOI: 10.25712/ASTU.2410-485X.2024.02) — EDN: https://elibrary.ru/HYHWHT
UDK 621.791
Research on Weld Pool Egmentation method for MAG Welding based on lightweight convolution Neural Network
авторы
1 School of Mechanical Engineering and Automation, Wuhan Textile University, Wuhan, 430073, China
E-mail: 19986541758@163.com
Abstract. During the welding process, the quality of the weld can be monitored by observing the morphology of the molten pool. However, interference such as arc light and spatter can affect the extraction of the molten pool morphology. This paper proposes a method for molten pool segmentation based on a lightweight convolutional neural network. First, a molten pool image acquisition platform was set up, and 292 images were collected. Secondly, a lightweight fully convolutional neural network (FCN) and a Deep Feature Aggregation Network (DFANet) were constructed. The experimental results show that among the FCN models, FCN-8 has the best segmentation performance. After using the Deep Feature Aggregation Neural Network, its segmentation performance was further improved. The accuracy of this network is 99.88%, the accuracy of the molten pool category is 92.17%, the intersection over union (IoU) for the molten pool is 85.89%, and the average IoU is 92.89%. Compared to FCN-8, it has improved by 3.37% in the average IoU metric. The use of DFANet for molten pool morphology extraction has practical promotion value.
Keywords: deep learning; deep aggregation neural network; full convolution neural network; molten pool
жтейдокя&н&ю MAG »тшшяшя
1 шхт^ шш^ш, йж, 430073
E-mail: 19986541758@163.com
Ш: ЖщМ+ЖЙЩШ«^, ШЮШШЙ, лтт
ш^ттттшшм. ij
ш, тттшштшш¥&, шт 292
(FCN) тШШШ^ШШШ (DFANet); ^МВШМШ
шхмтшшй^шт шшш, fcn fcn-8 ятштм, ш шш'ш&штш, шшшш-шя, шшжттъ
99.88%, ШШШШШ^ 92.17%, Шй^Ь^ 85.89%,
92.89%; Ш FCN-8 3.37%, ШШД DFANet ШШШ
ШП: ШВД; ШШ&ШШ;
0
тттхтАхттмшш, [1] ш
ж. ^нш^тшш, шшшш, шхттттх^х^ ±тшшх. ж^чяшшт^ш^, чтнтшттт^шш шм, шшшштшшшшм. ^шт тптх^ш, тштшш.
1.1 MAGjSffiX^fô^
mag ^ [2] тшхшт, ^«птешш^шй^, ш+ш
Ш&Ъ 12mm, 20# Mt^W, V M, V ШШШХШ
14mm, (Gas Metal Arc Welding,
GMAW), 600mm ЙЗД^ШтШШ,
птшх&штшхтп-ш, ш ^ mag шт^ш.
и
MAG
/Jn^
Figure 1. Schematic diagram of MAG welding pool
1.2 ШШ
Labelme Ш, # Python ЩЙФШЛ Labelme, тЖ
■ lAcl^c - □ X
tilt- E.-IH '¿е* №1р
\ J IK 9 к
. р-п
Ur-n [iL,
bbillirf » к
[>ЦУ
1141 l'ü^üür Lißtll 9 к
©
FdfLltt 9 к
SMlih rilfinpr
Iflbdme i-Urted.
Labelme^ffi Figure 2. LabelMe Interface
ш^штыш йлшш-ш, шшшыш, ^ашшшэ т
Ж, JPG+JSON т^, ШШ 6:2:2 Ш^УЯ^ШШ, ШШ
тшш, шш 292
щэ
Figure Э: Single Object Annotation Diagram
2 fcn ятт&яш
FCN (Fully Convolutional Networks, ±ШШМШ) ШЖШШШХЯШ1 [Э], ^
ш (fcn) щщ
ЖЙЙ, штжшшшш
ттШШ^гкШШМ. МТ-ШЯ^ CNN 1Ш, VGG т Resnet, ffi&ftm
^ softmax шшш^ш^м. шш
FCN-8 FCN-16 FCN-32
Figure 5. Comparison of network segmentation performance under different upsampling structures
®5 [5] FCN-8
ftm&MW^rn&mmi fcn-16 fcn-32 m^, bp fcn-8 m, (Miou), ^^
tt (Intersection over Union, IoU), (Class Pixel Accuracy, CPA), X
mm (Accuracy) mW^mU^K&W^ FCN ^ 1
FCN-8
Table 1. Comparison of Network Segmentation Performance
with Different Upsampling Structures
ACC (%) CPA (%) IoU (%) MIoU (%)
FCN-8 99.84 90.28 79.21 89.52
FCN-16 99.78 84.57 75.42 87.60
FCN-32
99.75
84.92
72.69
86.21
3 DFA'JJM^M
DFANet, gp Deep Feature Aggregation, ^E^^tT^
DFANet vttmm^xmmm^PT 8 ^it mm^T 2 tt ENet[6] i^mmm rnm^n^,
RRuminmrnrn^. s^, xception ^n fc a
attention, /AWM^feSf,
(DFA)
Ml^Xt DFANet DFANet (Deep Feature
Aggregation) IWS&
m R^T DFANet
m mmrn^mmmpFA) frm&m
Table 2. Segmentation Performance of Deep Aggregation
Neural Network (DFA)
ACC (%) CPA (%) IoU (%) MIoU (%)
DFA 99.88 92.17 85.89 92.89
Figure 6. Segmentation results of the molten pool by DFANet
4 mMrn
DFANet ^ttt^m "fft, FCN-8
m&fäT FCN-16, FCN-32, ÄÄ CPA
5.71% ^ 5.36%, Ä MIOU 1.92% ^ 3.31%,
^ftftft, FCN-8 IM^ÖÖ^f'Jtt
Ä&, DFANet ^fiJttf^Mt, DFANet IM^
FCN-8, ÄÄ CPA FCN-8 IM^Ä 1.89%, Ä MIOU
FCN-8 Ä 3.37%. ^Xttt, DFANet
DFANet IM^lMÄÄ'h -MÄäBffi®!, it
im^i^iM»^,
BISNET [7], STDC IM^IM,
it, ä
-^fött&Ätt CNN №ÄÄÄ MAG ^^fiJ + Ö«
ÄÄ^Ä^fe^Ä MAG MMMÖRTOÄ
[1] ?L«, «P, m. I^ttft^, 2024, 45(08):64-68. D01:10.16086/j.cnki.issn1000-0380.2022080090.
[2] ifät, I^Äfg. MAG XÄfrJf^, 2024, 37(09):110-112.
[3] ^gffi, m.
[J/OL]. f^^Ä^'J^^^, 1-16[2024-10-27]. https://doi.Org/10.13196/j.cims.2023.0729.
[4] 3BM&, i^m Ä m. FCN
2024, 40(04):5-10. D0I:10.20171/j.cnki.23-1419/n.2024.04.011.
[5] wää, mmm, m. ÄÄ yolov8 a^wfe
2024, 47(20):141-147. D0I:10.16652/j.issn.1004-373x.2024.20.022.
[6] mm ^M^, m. Ä«Ä ENet ^ißöx^^mtßö^^w^m.
M^PiH^fe^, 2023, 19(17):98-100. D0I:10.14004/j.cnki.ckt.2023.0927.
[7] ^mm, ^M, m. SX BiSNet ööMA^mitx^ii^em. MM
2021, 42(02):22-28. D0I:10.13435/j.cnki.ttc.003123. References
[1] Kong Weicheng, Sun Han, Li Silin, et al. Design of Local Autonomous Intelligent Welding Robot System [J]. Automation Instrumentation, 2024, 45(08): 64-68. DOI: 10.16086/j.cnki.issn1000-0380.2022080090.
[2] Wang Zhizhong, Chen Wentao. Analysis of Electrical Signals in MAG Welding Process [J]. Industrial Control Computer, 2024, 37(09): 110-112.
[3] Han Songjie, Liu Yinhua, Li Yanzheng, et al. Semi-supervised Semantic Segmentation Method for Hig-h-density Point Cloud Data in Car Body Welding Scenario [J/OL]. Computer Integrated Manufacturing Sy-stems, 1-16 [2024-10-27]. https://doi.org/10.13196/j.cims.2023.0729.
[4] Zou Xiand, He Xiaoli, Yu Qian, et al. An Improved FCN-based Method for Pneumonia Image Segme-ntation [J]. Journal of Qiqihar University (Natural Science Edition), 2024, 40(04): 5-10. DOI: 10.20171/j.cnki.23-1419/n.2024.04.011.
[5] Zhao Huipeng, Cao Jingsheng, Pan Dijin, et al. Small Target Detection of Traffic Signs with Improve-d YOLOv8 Algorithm [J]. Modern Electronic Technology, 2024, 47(20): 141-147. DOI: 10.16652/j.issn.1004-373x.2024.20.022.
[6] Lian Bobo, Zang Mengte, Fu Xianjun, et al. Foreign Object Detection Method in Industrial Belt Conv-eyor Based on Improved ENet Algorithm [J]. Computer Knowledge and Technology, 2023, 19(17): 98-100. DOI: 10.14004/j.cnki.ckt.2023.0927.
[7] Li Pengcheng, Chen Haidong, Li Shipeng, et al. Intelligent Health Monitoring of Aerospace Engine R-otor Based on BiSNet [j]. Remote Sensing and Control, 2021, 42(02): 22-28. DOI: 10.13435/j.cnki.ttc.003123.
