基于动态权重的自适应PSO-BP神经网络焊接缺陷成因诊断

高昶霖; 宋燕利; 左洪洲; 章诚

doi:10.12073/j.hjxb.20210515001

基于动态权重的自适应PSO-BP神经网络焊接缺陷成因诊断

高昶霖^1,,
宋燕利^{1, 2, 3, ,},
左洪洲⁴,
章诚¹

1.
武汉理工大学, 现代汽车零部件技术湖北省重点实验室, 武汉, 430070
2.
武汉理工大学, 汽车零部件技术湖北省协同创新中心, 武汉, 430070
3.
武汉理工大学, 湖北省材料绿色精密成形工程技术研究中心, 武汉, 430070
4.
湖北省齐星汽车车身股份有限公司, 随州, 441300

基金项目: 湖北省技术创新专项(2019AAA014)；湖北省重点研发计划项目(2020BAB143)；新能源汽车科学与关键技术学科创新引智基地(B17034)；教育部创新团队发展计划资助项目(IRT_17R83)

详细信息

作者简介:
高昶霖，硕士研究生；主要研究方向为专家系统、智能制造；Email：506070302@qq.com

通讯作者:
宋燕利，博士，教授，博士研究生导师；Email：ylsong@whut.edu.cn.

中图分类号: TG409
计量
- 文章访问数: 364
- HTML全文浏览量: 43
- PDF下载量: 43
出版历程
- 收稿日期: 2021-05-14
- 录用日期: 2022-02-14
- 网络出版日期: 2022-02-18
- 刊出日期: 2022-01-24

Cause diagnosis of welding defects based on adaptive PSO-BP neural network with dynamic weighting

GAO Changlin^1,,
SONG Yanli^{1, 2, 3, ,},
ZUO Hongzhou⁴,
ZHANG Cheng¹

1.
Hubei Key Laboratory of Advanced Technology for Automotive Components, Wuhan University of Technology, Wuhan, 430070, China
2.
Hubei Collaborative Innovation Center for Automotive Components Technology, Wuhan University of Technology, Wuhan, 430070, China
3.
Hubei Engineering Research Center for Green & Precision Material Forming, Wuhan University of Technology, Wuhan, 430070, China
4.
Hubei Qixing Automobile Body Co., Ltd., Suizhou, 441300, China

摘要

摘要: 焊接缺陷产生原因复杂，影响因素众多，基于人工智能的缺陷成因诊断算法成为焊接智能化的发展方向. 将PSO-BP神经网络应用于焊接缺陷成因诊断，利用神经网络的连接学习机制代替传统专家系统的规则推理机制，并对PSO算法进行自适应调整，引入动态权重因子，搭建自适应PSO-BP神经网络模型. 结果表明，与传统PSO-BP神经网络模型相比，改进的PSO-BP神经网络模型训练所需要的迭代次数减少13.1%，诊断结果准确率从93.3%提高至96.7%，精确率从91.3%提高至98.3%，综合能力指标从91.7%提高至96.9%. 改进算法能够明显提升焊接缺陷成因诊断的效率和精度，具有较好的工程应用价值.
- 焊接缺陷成因 /
- 神经网络 /
- 动态权重 /
- 自适应
Abstract: Considering the complex causes and various impact factors for welding defects, diagnosis methods based on artificial intelligence algorithms are regarded as one of the directions for the development of intelligentizing welding. In this study, an improved diagnosis method for welding defect based on PSO-BP neural network is proposed. Connection learning mechanism of neural network is used instead of the rule reasoning mechanism of traditional expert systems. It also makes adaptive adjustments to the PSO algorithm, introduced dynamic weight factors, and builds an adaptive PSO-BP neural network model. Compared with the traditional PSO-BP neural network model, the number of iterations required to train the improved PSO-BP neural network model reduced by 13.1%, the accuracy of diagnostic results increased from 93.3% to 96.7%, the precision increased from 91.3% to 98.3%, and the comprehensive performance index increased from 91.7% to 96.9%. The results show that the improved algorithm can significantly improve the efficiency and accuracy of welding defect diagnosis, and has good engineering application value.
- cause of welding defects /
- neural network /
- dynamic weighting /
- adaptive

HTML全文

图 1 不同算法流程

Figure 1. Different algorithms flow. (a) traditional PSO-BP algorithm; (b) adaptive PSO-BP algorithm based on dynamic weighting

下载: 全尺寸图片幻灯片

图 2 焊接缺陷成因智能诊断算法流程

Figure 2. Intelligent diagnosis algorithm process for welding defect causes

下载: 全尺寸图片幻灯片

图 3 BP神经网络焊接缺陷分析模型

Figure 3. BP neural network model for welding defect analysis

下载: 全尺寸图片幻灯片

图 4 网格搜索法确定隐藏层神经元个数

Figure 4. Grid search to determine the number of neurons in the hidden layer

下载: 全尺寸图片幻灯片

图 5 损失函数曲线

Figure 5. Loss function curve

下载: 全尺寸图片幻灯片

图 6 不同算法神经网络损失函数曲线

Figure 6. Loss function curve of neural networks with different algorithms

下载: 全尺寸图片幻灯片

图 7 缺陷形成原因混淆矩阵对比

Figure 7. Comparison of confusion matrix of defect causes. (a) adaptive PSO-BP algorithm based on dynamic weighting; (b) traditional PSO-BP algorithm; (c) BP neural network

下载: 全尺寸图片幻灯片

图 8 参数调整前后焊缝形貌

Figure 8. Weld morphology before and after parameter adjustment. (a) before parameter adjustment; (b) after parameter adjustment

下载: 全尺寸图片幻灯片

表 1 基于MIV的焊接成因诊断输入特征变量灵敏度分析

Table 1 Sensitivity analysis of input characteristic variables for welding cause diagnosis based on MIV

焊接参数	平均影响值 δ_IV	各焊接参数MIV 所占的比例η（%）
缺陷种类C	0.637 5	26.042
缺陷相对中心距离l₁/mm	0.060 5	2.471
两缺陷最近距离l₂/mm	0.029 8	1.217
缺陷长宽比δ	−0.207 3	8.468
板厚b/mm	0.051 3	2.096
焊接电流I/A	0.337 2	13.775
焊接电压U/V	0.570 3	23.296
焊接速度v/(mm·s⁻¹)	−0.179 7	7.341
送丝速度v_s/(m·min⁻¹)	0.139 9	5.715
气体流量Q/(L·min⁻¹)	−0.234 5	9.579

下载: 导出CSV

表 2 基于不同算法的焊接缺陷成因诊断结果

Table 2 Diagnosis results of welding defects based on different algorithms

神经网络种类	加权平均评价指标
神经网络种类	准确率A_CC(%)	精确率 P(%)	召回率 R(%)	综合能力指标F₁(%)
基于动态权重的自适应PSO-BP	96.7	98.3	96.7	96.9
传统PSO-BP	93.3	91.3	93.3	91.7
BP	90.0	88.0	90.0	88.6

下载: 导出CSV

表 3 测试样本的输入特征

Table 3 Input features of the test samples

样本编号	缺陷种类 C	缺陷长宽比 δ	焊接电流 I/A	焊接电压 U/V	焊接速度 v/(mm·s⁻¹)	送丝速度 v_s/(m·min⁻¹)	气体流量 Q/(L·min⁻¹)
1	气孔	1.10	171.5	19.2	10.90	9.30	18
2	焊穿	8.81	180.0	20.6	9.33	10.00	20
3	未焊透	7.33	126.0	19.5	6.67	8.00	20
4	焊穿	6.49	143.5	19.6	4.00	8.00	19
5	裂纹	10.45	178.0	19.2	9.67	9.30	19

下载: 导出CSV

表 4 缺陷成因真实值及网络输出结果

Table 4 Real value of defect causes and network output results

输出特征	真实值T(%)					网络输出值P_o(%)
输出特征	样本1	样本2	样本3	样本4	样本5	样本1	样本2	样本3	样本4	样本5
非工艺参数原因	0	0	0	0	0	0	0	0	0	0
焊接电流过大	0	0	0	0	100	0	0	0	0	96.61
焊接电流过小	0	0	100	0	0	0	0	96.11	2.58	2.15
焊接电压过大	0	100	0	0	0	0	95.08	0	3.32	0
焊接电压过小	0	0	0	0	0	0	0	0	0	0
焊接速度过大	100	0	0	0	0	99.51	2.75	1.46	0	0
焊接速度过小	0	0	0	100	0	0	0	2.43	94.10	0
送丝速度过大	0	0	0	0	0	0	0	0	0	0
送丝速度过小	0	0	0	0	0	0	0	0	0	0
气体流量过大	0	0	0	0	0	0.38	0	0	0	1.24
气体流量过小	0	0	0	0	0	0.11	2.17	0	0	0

下载: 导出CSV

参考文献(17)

[1]	吴叶军, 魏艳红. 智能化焊接CAPP的分析与开发[J]. 焊接学报, 2015, 36(7): 109 − 112. Wu Yejun, Wei Yanhong. Analysis and development of intelligentialized welding CAPP system[J]. Transactions of the China Welding Institution, 2015, 36(7): 109 − 112.
[2]	Tsoukalas V D, Kontesis M, Badogiannis E, et al. WELDES: An intelligent defects expert system for aluminum welding process[J]. Wseas Transactions on Information Science & Applications, 2007, 4(2): 339 − 345.
[3]	宋燕利, 余成, 戴定国, 等. 基于BP 和GA 的激光焊接热源模型参数优化[J]. 塑性工程学报, 2017, 24(1): 218 − 222. Song Yanli, Yu Cheng, Dai Dingguo, et al. Parameter optimization of heat source model for laser welding based on BP neural network and genetic algorithm[J]. Journal of Plasticity Engineering, 2017, 24(1): 218 − 222.
[4]	邵晴, 于庆斌, 尹华, 等. 焊接热输入对高速动车组转向架侧梁焊接变形的影响及优化[J]. 焊接学报, 2020, 41(12): 25 − 32,48. doi: 10.12073/j.hjxb.20200216002 Shao Qing, Yu Qingbin, Yin Hua, et al. Effect of welding heat input on welding deformation of bogie side beam of high-speed EMU and optimization[J]. Transactions of the China Welding Institution, 2020, 41(12): 25 − 32,48. doi: 10.12073/j.hjxb.20200216002
[5]	Liu J, Xu G C, Ren L, et al. Defect intelligent identification in resistance spot welding ultrasonic detection based on wavelet packet and neural network[J]. The International Journal of Advanced Manufacturing Technology, 2017, 90(9-12): 2581 − 2588. doi: 10.1007/s00170-016-9588-y
[6]	Li Z K, Zhao X H. BP artificial neural network based wave front correction for sensor-less free space optics communication[J]. Optics Communications, 2017, 385: 219 − 228. doi: 10.1016/j.optcom.2016.10.037
[7]	刘占军, 单宝峰, 贺平. 小波神经网络在铝合金焊接缺陷诊断中的研究[J]. 振动、测试与诊断, 2005, 25(3): 219 − 221. doi: 10.3969/j.issn.1004-6801.2005.03.013 Liu Zhanjun, Shan Baofeng, He Ping. Wavelet neural network application to diagnosing defect of aluminum alloy welding[J]. Journal of Vibration, Measurement & Diagnosis, 2005, 25(3): 219 − 221. doi: 10.3969/j.issn.1004-6801.2005.03.013
[8]	姜洪权, 贺帅, 高建民, 等. 一种改进卷积神经网络模型的焊缝缺陷识别方法[J]. 机械工程学报, 2020, 56(8): 235 − 242. doi: 10.3901/JME.2020.08.235 Jiang Hongquan, He Shuai, Gao Jianmin, et al. An improved convolutional neural network for weld defect recognition[J]. Journal of Mechanical Engineering, 2020, 56(8): 235 − 242. doi: 10.3901/JME.2020.08.235
[9]	Ravi R, Aaquib R K, Chirag P, et al. Classification and identification of surface defects in friction stir welding: An image processing approach[J]. Journal of Manufacturing Processes, 2016, 22: 237 − 253. doi: 10.1016/j.jmapro.2016.03.009
[10]	Bacioiu D, Melton G, Papaelias M, et al. Automated defect classification of aluminium 5083 TIG welding using HDR camera and neural networks[J]. Journal of Manufacturing Processes, 2019, 45: 603 − 613. doi: 10.1016/j.jmapro.2019.07.020
[11]	Zhang Z, Jia L M, Qin Y. Modified constriction particle swarm optimization algorithm[J]. Journal of Systems Engineering and Electronics, 2015, 26(5): 1107 − 1113.
[12]	张爱华, 高佛来, 牛小革, 等. 基于BP神经网络的钢轨闪光对焊接头灰斑面积预测[J]. 焊接学报, 2016, 37(11): 11 − 14. Zhang Aihua, Gao Folai, Niu Xiaoge, et al. Prediction of gray-spot area in rail flash butt welded joint based on BP neural network[J]. Transactions of the China Welding Institution, 2016, 37(11): 11 − 14.
[13]	Avidan S. Ensemble tracking[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2007, 29(2): 261 − 271. doi: 10.1109/TPAMI.2007.35
[14]	Li J Y, Men C, Qi J F, et al. Impact factor analysis, prediction, and mapping of soil corrosion of carbon steel across China based on MIV-BP artificial neural network and GIS[J]. Journal of Soils and Sediments, 2020, 20(8): 3204 − 3216. doi: 10.1007/s11368-020-02649-5
[15]	刘碧瑶. 基于BP神经网络的住院费用建模研究[D]. 杭州: 浙江大学, 2006. Liu Biyao. Research of establishing hospitalization charge fitting model by using BP neural network[D]. Hangzhou: Zhejiang University, 2006.
[16]	Sheela K G, Deepa S N. Review on methods to fix number of hidden neurons in neural networks[J]. Mathematical Problems in Engineering, 2013(6): 425740.
[17]	王东风, 孟丽. 粒子群优化算法的性能分析和参数选择[J]. 自动化学报, 2016, 42(10): 1552 − 1561. Wang Dongfeng, Meng Li. Performance analysis and parameter selection of PSO algorithms[J]. Acta Automatica Sinica, 2016, 42(10): 1552 − 1561.

施引文献(6)

期刊类型引用(4)

1.	许峰，杨莉，熊义峰，刘坡，张尧成. 搅拌摩擦加工Al-Pb表面复合材料的微结构和织构. 稀有金属材料与工程. 2021(03): 957-962 . 百度学术
2.	张亚敏，姜永亮. 基于神经网络算法的铝基复合材料搅拌铸造工艺优化. 热加工工艺. 2021(18): 91-94 . 百度学术
3.	张孙艺，周爽，朱绍举，高吉成. 基于FSP技术制备CeO_2颗粒增强铝基复合材料. 机械工程与自动化. 2020(02): 151-152 . 百度学术
4.	杨绍斌，张旭，谢帅. 高质量分数Al_2O_3/Al复合材料的硬度和耐磨性能. 材料保护. 2018(04): 47-50+140 . 百度学术