Penetration Recognition and Close-loop Control Method of Fast-frequency Pulsed TIG Welding Process for 304 Stainless Steel
- Vol. 53, Issue 9, Pages: 47-54(2023)
DOI: 10.7512/j.issn.1001-2303.2023.09.06
扫 描 看 全 文
扫 描 看 全 文
李柳熠,陈浩宇,董荧,等.304不锈钢快频脉冲TIG焊的熔透识别与闭环控制[J].电焊机,2023,53(9):47-54.
LI Liuyi, CHEN Haoyu, DONG Ying, et al.Penetration Recognition and Close-loop Control Method of Fast-frequency Pulsed TIG Welding Process for 304 Stainless Steel[J].Electric Welding Machine, 2023, 53(9): 47-54.
焊缝熔透的识别与控制对提高快频脉冲TIG焊(Fast-frequency Pulsed Tungsten Inert Gas welding,FFP-TIG welding)的焊接质量和自动化水平具有重要意义。熔池图像可以作为卷积神经网络(Convolution Neural Network,CNN)的输入来识别熔透状态,但模型往往较为复杂,难以部署于工业边缘设备中,同时缺乏基于识别结果的熔透状态控制方法。针对该问题,搭建视觉传感系统来构建304不锈钢FFP-TIG焊熔池图像数据集,包括未熔透、适度熔透和过度熔透三种标签;开发基于知识蒸馏(Knowledge Distillation,KD)的熔透识别模型,利用MobileNetV2作为教师模型,用于训练自定义的学生模型ResNet-KD;基于ResNet-KD的softmax层输出向量设计模糊控制器,实时调整快频基值电流,实现熔透状态的闭环控制。结果表明,ResNet-KD在验证集上的准确率达到了97.60%,在预设电流、变散热及热量积累的工况下,模糊控制器均有良好的性能。
Weld penetration recognition and control is of great significance to improve welding quality and automation level of Fast-frequency Pulsed Tungsten Inert Gas welding (FFP-TIG welding). Weld pool images can be used as the input of convolutional neural network (CNN) to recognize the penetration state, but the model is often complex and difficult to deploy in industrial edge devices. At the same time, it lacks control methods for penetration state based on recognition results. To solve this problem, a visual sensing system was established to construct the FFP-TIG welding pool image dataset of 304 stainless steel, including lack of penetration, desirable penetration, and excessive penetration. A penetration recognition model based on knowledge distillation (KD) was developed, which utilized MobileNetV2 as the teacher model for training ResNet-KD that is a custom student model. Based on the softmax layer output of ResNet-KD, a fuzzy controller was designed to adjust the fast-frequency base current in real-time and realize closed-loop control of the penetration state. The results showed that the recognition accuracy of ResNet-KD on the validation set achieved 97.60%, and the fuzzy controller had good performance under the conditions of presetting initial cparameters, heat dissipation variation and heat accumulation.
快频脉冲TIG焊熔池图像熔透识别知识蒸馏模糊控制
FFP-TIG weldingweld pool imagepenetration recognitionknowledge distillationfuzzy control
吴健文, 徐孟嘉, 范文艳, 等. 钛合金快频脉冲柔性波形调制TIG焊接工艺[J]. 机械工程学报, 2020, 56(06): 102-109.
WU J W,XU M J,FAN W Y,et al. Flexible Waveform Interpulse TIG Welding for Titanium Alloy[J]. Journal of Mechanical Engineering, 2020, 56(06): 102-109.
Zhang Z, Wen G, Chen S. Weld image deep learning-based on-line defects detection using convolutional neural networks for Al alloy in robotic arc welding[J]. Journal of Manufacturing Processes,2019,45:208-216.
梁明龙,周凯荣,倪加明,等. 2219铝合金变极性钨极惰性气体保护焊的焊缝成形模糊控制系统[J]. 上海交通大学学报,2016,50(10):1583-1587.
LIANG M L,ZHOU K R,NI J M,et al. Development of a Variable Polarity Gas Tungsten Arc Welding Fuzzy Control System for 2219 Aluminium Alloy[J]. Journal of Shanghai Jiaotong University,2016,50(10):1583-1587.
Cheng Y, Chen S, Xiao J, et al. Dynamic estimation of joint penetration by deep learning from weld pool image[J]. Science and Technology of Welding and Joining, 2021, 26(4): 279-285.
Zhang B, Shi Y, Cui Y, et al. Prediction of keyhole TIG weld penetration based on high-dynamic range imaging[J]. Journal of Manufacturing Processes, 2021, 63: 179-190.
Chen C, Lv N, Chen S. Welding penetration monitoring for pulsed GTAW using visual sensor based on AAM and random forests[J]. Journal of Manufacturing Processes, 2021, 63: 152-162.
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.
LI H C,LIU J F,XIE J B,et al. GTAW Penetration Prediction Model Based on Convolution Neural Network Algorithm[J]. Journal of Mechanical Engineering, 2019, 55(17): 22-28.
Jiao W, Wang Q, Cheng Y, et al. End-to-end prediction of weld penetration: A deep learning and transfer learning based method[J]. Journal of Manufacturing Processes, 2021, 63: 191-197.
Wang Z, Chen H, Zhong Q, et al. Recognition of penetration state in GTAW based on vision transformer using weld pool image[J]. International Journal of Advanced Manufacturing Technology, 2022, 119(7-8): 5439-5452.
Liu T,Wang J,Huang X, et al. 3DSMDA-Net: An improved 3DCNN with separable structure and multi-dimensional attention for welding status recognition[J].Journal of Manufacturing Systems,2022, 62:811-822.
吴頔. 基于多源信息融合的铝合金VPPAW成形预测和智能控制研究[D]. 上海:上海交通大学, 2018.
WU D. Research on Predicting and Intelligent Control for Weld Formation During VPPAW Process Using Multi-information Fusion[D]. Shanghai: Shanghai Jiao Tong University, 2018.
燕旭. 熔孔型TIG单面焊双面成形视觉检测及控制技术研究[D]. 山东:山东大学, 2021.
YAN X. Study on Visual Inspection and Control Technology of fusion hole TIG single side welding with back formation[D]. Shandong:Shandong University,2021.
Wang Q, Jiao W, Zhang Y M. Deep learning-empow-ered digital twin for visualized weld joint growth monitoring and penetration control[J]. Journal of Manufacturing Systems, 2020, 57: 429-439.
编辑部网址:http://www.71dhj.comhttp://www.71dhj.com
相关文章
相关作者
相关机构