基于SSTFT与KSVD的异种材料FSW在线监测

龙海威; 张佳莹; 刘锐; 孙屹博; 魏晓; 杨鑫华

doi:10.12073/j.hjxb.20240716002

基于SSTFT与KSVD的异种材料FSW在线监测

1.
大连交通大学, 詹天佑学院, 大连, 116028
2.
特种车辆设计与制造集成技术国家重点实验室, 包头, 014032
3.
辽宁省轨道交通装备焊接与可靠性重点实验室, 大连, 116028

基金项目: 国家自然科学基金资助项目(52005071)；特种车辆设计与制造集成技术国家重点实验室(GZ2023KF015)；辽宁省自然科学基金(2022-MS-342)

详细信息

作者简介:
龙海威，博士研究生；主要研究方向为焊接过程检测与控制；Email: 1172527831@qq.com

通讯作者:
孙屹博，博士，教授，博士研究生导师；Email: yibo_sun@126.com.

中图分类号: TG 441.7
计量
- 文章访问数: 68
- HTML全文浏览量: 9
- PDF下载量: 17
出版历程
- 收稿日期: 2024-07-15
- 网络出版日期: 2024-09-17
- 刊出日期: 2024-11-24

Online monitoring of dissimilar material FSW based on SSTFT and KSVD

1.
College of China Railway Rolling Stock Corporation, Dalian Jiaotong University, Dalian, 116028, China
2.
State Key Laboratory of Special Vehicle Design and Manufacturing Integration Technology, Baotou, 014032, China
3.
Dalian Key Laboratory of Welded Structures and Intelligent Manufacturing Technology of Rail Transportation Equipment, Dalian, 116028, China

摘要

摘要:
异种材料轻量化结构是航空航天、铁路、汽车等领域的关键技术和研究热点之一，搅拌摩擦焊(FSW)是连接异种材料的有效方法，由于异种材料物理和化学性质的不同，容易在焊接过程中产生缺陷. 针对铝合金与碳纤维增强热塑性塑料(CFRTP)搅拌摩擦焊(FSW)缺陷监测提出了基于同步压缩短时傅立叶变换与K-奇异值分解(SSTFT-KSVD)在线监测方法. 使用声发射(AE)信号实时监测FSW状态，利用同步压缩短时傅立叶变换(SSTFT)提取时频域特征，最后通过K-奇异值分解(KSVD)模型对焊接状态与焊接缺陷进行了分类. 结果表明，AE信号频率成分集中在10 kHz，17 kHz，23 kHz和25 kHz 4个频段，熔核塌陷和表面擦伤缺陷发生时，23 kHz频段的功率分别转移到10 kHz，而表面擦伤发生时，25 kHz频段的功率转移到17 kHz. 在缺陷预测方面，KSVD预测模型的平均准确率达到90%，响应时间达到10 ms量级，比神经网络快100倍. 基于SSTFT-KSVD在线监测方法可以实现对Al-CFRTP异种材料的FSW快速监测.
- 异种材料搅拌摩擦焊 /
- 声发射信号 /
- 同步压缩短时傅立叶变换 /
- K-奇异值分解 /
- 在线监测
Abstract:
Lightweight is one of the key technologies and research hotspots in the fields of aerospace, railway, and vehicles, which makes the demand for the joining of dissimilar materials such as Al-based and carbon fiber-based materials. Friction stir welding (FSW) is a potential method for the joining of two kinds of materials, however, defects are easily generated due to different physical and chemical properties. In this paper, an in-situ monitoring and defect diagnosis method for FSW of 6061-T6 Al alloy and Continuous Fiber reinforced thermoplastic plastics (Al-CFRTP) is proposed. In this method, the acoustic emission (AE) signal at the upper workpiece of the lap joint is obtained. Time-frequency domain features are extracted by synchronous compression short-time Fourier transform (SSTFT). The welding status is classified by the K-singular value decomposition (KSVD), including pin tool pressure and down, nugget collapse defects, flash defects, and surface galling defects. The Al-CFRTP FSW experiments are carried out based on this method. From the SSTFT results, the dominant frequency of AE signals is concentrated in four main frequency bands of 10 kHz, 17 kHz, 23 kHz, and 25 kHz. The frequency band shifts from 23 kHz to 10 kHz accompanied by the nugget collapse defects and shifts from 25 kHz to 17 kHz accompanied by the flash and sueface galling defects. Results indicate that the recognition accuracy of defects reaches 90% based on KSVD prediction model. The computation speed is 100 times faster than neural networks with nearly the same recognition precision. The SSTFT-KSVD method is effective for In-situ monitoring and defect diagnosis of the FSW process.
- FSW of dissimilar materials /
- AE signal /
- synchronous compression short-time Fourier transform /
- K-singular value decomposition /
- online monitoring

HTML全文

图 1 FSW声发射在线监测示意图

Figure 1. Diagram of FSW AE online monitoring. (a) experimental equipment and signal acquisition platform for FSW; (b) schematic diagram of AE sensor

下载: 全尺寸图片幻灯片

图 2 SSTFT-KSVD缺陷预测方法的基本流程图

Figure 2. Basic flow chart of SSTFT-KSVD defect prediction method

下载: 全尺寸图片幻灯片

图 3 4组试验焊接结果与声发射信号

Figure 3. Welding results and acoustic emission signals of four experiments. (a) experiments of No.1; (b) experiments of No.2; (c) experiments of No.3; (d) experiments of No.4

下载: 全尺寸图片幻灯片

图 4 4组试验SSTFT分析结果

Figure 4. Results of SSTFT in four experimental. (a) experiments of No.1; (b) experiments of No.2; (c) experiments of No.3; (d) experiments of No.4

下载: 全尺寸图片幻灯片

图 5 谱相关聚类结果

Figure 5. Results of spectral correlation clustering. (a) scatter diagram of spectral correlation clustering; (b) welding stage in experiment of No.3

下载: 全尺寸图片幻灯片

图 6 SSTFT-KSVD缺陷预测结果

Figure 6. Result of the SSTFT-KSVD defect prediction. (a) experiments of No.1; (b) experiments of No.2; (c) experiments of No.3; (d) experiments of No.4

下载: 全尺寸图片幻灯片

表 1 铝合金和CFRTP的力学性能

Table 1 Mechanical properties of Al alloy and CFRTP

材料	抗拉强度 R_m/MPa	导热系数 γ/(W·m⁻¹·K⁻¹)	初熔温度 T/℃
6061-T6	260	167	652
CFRTP	53	4.42	280

下载: 导出CSV

表 2 FSW工艺参数

Table 2 Experiment parameters of FSW

试验编号	焊接速度 v/(mm·min⁻¹)	搅拌头转速 n/(r·mim⁻¹)	搅拌头下压量 h/mm
1	10	1 800	2.5
2	15	1 800	2.5
3	20	1 800	2.5
4	25	1 800	2.5

下载: 导出CSV

表 3 不同字典维度对字典学习的影响

Table 3 Influence of different dictionary dimensions on dictionary learning

字典维度	训练次数 N₁	稀疏惩罚系数 a(10⁻⁶)	均方根误差 R_RMSE (10⁻¹⁸)
4 × 4	217	6.2	13.8
4 × 8	304	5.8	8.6
4 × 10	344	5.1	3.1
4 × 12	383	6.1	8.5
4 × 16	461	6.5	14.2

下载: 导出CSV

表 4 混淆矩阵的评价指标.

Table 4 Evaluation index of the confusion matrix.

类别	准确率P	召回率R	F1-Score评价指标F
噪声区	0.925	0.923	0.924
搅拌头动作区	0.900	0.898	0.898
缺陷区(熔核塌陷缺陷)	0.898	0.896	0.897
缺陷区(表面擦伤缺陷)	0.905	0.910	0.907

下载: 导出CSV

表 5 神经网络的比较

Table 5 Comparison of multiple neural networks

分类算法	平均准确率 mP(%)	迭代次数 N₂	计算时间 t(毫秒级)
BP	92.1	593	100
RBF	92.5	675	100
ResNet	93.3	832	1 000
Res-GCM ^[14]	95.6	1 357	1 000
KSVD(本文提出)	90.7	344	10

下载: 导出CSV

参考文献(14)

[1]	Esteves J, Goushegir S, Santos J, et al. Friction spot joining of aluminum AA6181-T4 and carbon fiber-reinforced poly (phenylene sulfide): Effects of process parameters on the microstructure and mechanical strength[J]. Materials & Design, 2015, 66: 437 − 445.
[2]	朱浩, 张二龙, 莫淑娴, 等. 带状组织对铝/镁异种金属搅拌摩擦焊接头力学性能的影响[J]. 焊接学报, 2020, 41(1): 34 − 38. Zhu Hao, Zhang Erlong, Mo Shuxian, et al. Effect of banded microstructure on the mechanical properties of aluminum/magnesium dissimilar metal stir friction welded joints[J]. Transactions of the China Welding Institution, 2020, 41(1): 34 − 38.
[3]	Zhu L, Li N, Childs P. Light-weighting in aerospace component and system design[J]. Propulsion and Power Research, 2018, 7(2): 103 − 119. doi: 10.1016/j.jppr.2018.04.001
[4]	金启豪, 陈娟, 彭立明, 等. 碳纤维增强树脂基复合材料与铝 / 镁合金连接研究进展[J]. 材料工程, 2022, 50(1): 15 − 24. doi: 10.11868/j.issn.1001-4381.2021.000524 Jin Qihao, Chen Juan, Peng Liming, et al. Research progress on the connection between carbon fiber reinforced resin based composites and aluminum/magnesium alloys[J]. Materials Engineering, 2022, 50(1): 15 − 24. doi: 10.11868/j.issn.1001-4381.2021.000524
[5]	Sun Y, Zhang Y, Long H, et al. Friction stir lap welding for dissimilar materials of aluminum alloy and carbon-fiber-reinforced polyetherimide[J]. Materials Today Communications, 2022, 33: 104427. doi: 10.1016/j.mtcomm.2022.104427
[6]	Sen M, Chattopadhyaya S. Investigations into FSW joints of dissimilar aluminum alloys[J]. Materials Today: Proceedings, 2020, 27: 2455 − 2462. doi: 10.1016/j.matpr.2019.09.218
[7]	Dmitriev A, Polyakov V, Kolubaev E. Diagnostics of aluminum alloys with friction stir welded joints based on multivariate analysis of acoustic emission signals[J]. Journal of Physics: Conference Series. IOP Publishing, 2020, 1615(1): 012003. doi: 10.1088/1742-6596/1615/1/012003
[8]	Dejans A, Kurtov O, Rymenant P. Acoustic emission as a tool for prediction of nugget diameter in resistance spot welding[J]. Journal of Manufacturing Processes, 2021, 62: 7 − 17. doi: 10.1016/j.jmapro.2020.12.002
[9]	Li M, Ma C, Dang W, et al. DSCNN: dilated shuffle CNN model for SSVEP signal classification[J]. IEEE Sensors Journal, 2022, 22(12): 12036 − 12043. doi: 10.1109/JSEN.2022.3173433
[10]	Yaacoubi S, Dahmene F, Mountassir M, et al. A novel AE algorithm-based approach for the detection of cracks in spot welding in view of online monitoring: case study[J]. The International Journal of Advanced Manufacturing Technology, 2021, 117(5-6): 1807 − 1824. doi: 10.1007/s00170-021-07848-z
[11]	刘清泉, 张亚飞, 李华锋, 等. 基于协作低秩分层稀疏和LC-KSVD的人脸表情识别[J]. 传感器与微系统, 2017, 36(11): 56 − 59. Liu Qingquan, Zhang Yafei, Li Huafeng, et al. Facial expression recognition based on collaborative low rank hierarchical sparsity and LC-KSVD[J]. Sensors and Microsystems, 2017, 36(11): 56 − 59.
[12]	张利平, 邵宗凯, 吴建德. 基于改进KSVD和极限学习机的车型识别方法研究[J]. 计算机与数字工程, 2016, 44(6): 1093 − 1096. Zhang Liping, Shao Zongkai, Wu Jiande. Research on vehicle identification method based on improved KSVD and extreme learning machine[J]. Computer and Digital Engineering, 2016, 44(6): 1093 − 1096.
[13]	郑小磊. 基于增强的时频分析方法的变转速轴承故障诊断[J]. 轴承, 2024, 3(5): 1 − 9. Zheng Xiaolei. Fault diagnosis of variable speed bearings based on enhanced time-frequency analysis method[J]. Bearing, 2024, 3(5): 1 − 9.
[14]	Li X, Shi Y, Jian Y, et al. Research on welding penetration status monitoring based on Residual-Group convolution model[J]. Optics & Laser Technology, 2023, 163: 109322.