点焊质量诊断模型的泛化性能研究及改进

doi:10.3901/JME.2025.18.064

摘要/Abstract

摘要： 电阻点焊是薄壁结构制造的核心基础工艺，利用焊中传感信号实现点焊质量在线诊断是保障产品可靠性、推动制造模式转型升级的重要途径。然而，大批量生产中的装配偏差会导致焊接工况随机波动，影响接头成形质量，如何提高点焊质量诊断模型由正常工况向异常工况的外推泛化能力，是提升在线检测可靠性的关键。建立多种工况下的点焊质量与多传感信号数据集，提出一种基于自编码器结构和域对抗训练的传感信号特征提取方法。通过自编码器将时序过程信号压缩为高维特征空间中的特征向量，并结合注意力机制，自适应挖掘过程信号中的关键时间步信息，显著增强了特征表达能力。通过域对抗训练缩小不同工况下信号特征的分布差异，提高了跨工况质量诊断精度。分别采用正常工况和异常工况下的数据集作为训练和测试样本，对比所提出的新方法和人工方法以及三种常用的特征提取算法(PCA、Isomap、LLE)在焊点质量诊断任务上的工况泛化效果。结果表明，与其余四种方法相比，新方法在异常工况测试集上的焊点质量诊断准确率从83.18%提升至93.64%，熔核直径预测的均方根误差减小近37.5%，可有效提高质量诊断模型对分布外数据的外推泛化能力。

关键词: 电阻点焊, 质量在线诊断, 注意力机制, 域对抗网络, 分布外泛化

Abstract: Resistance spot welding (RSW) is a fundamental process in thin-walled structure manufacturing. Utilizing process signals during welding to achieve online quality diagnosis of the welds is a crucial approach to ensuring product reliability and advancing the transformation and upgrade of manufacturing modes. However, assembly deviations in mass production can lead to random fluctuations in welding conditions, impacting the quality of joint formation. Enhancing the generalization ability of the quality diagnosis model from normal conditions to abnormal ones is the key to improving the reliability of online detection. In this paper, a dataset of weld quality label under various conditions and multi-sensor signal data is established. A feature extraction method for sensor signals based on an autoencoder and domain adversarial training is proposed. The process signal is compressed into a feature vector in a high-dimensional feature space through the autoencoder, and the extraction of features with strong expressive ability is realized by combining the attention mechanism. Domain adversarial training is then utilized to reduce distributional differences in signal features across different conditions, thereby improving cross-condition quality diagnosis accuracy. Data from normal and abnormal welding conditions are used as training and test samples, respectively, to compare the generalization performance of the proposed method against manual methods and three commonly used feature extraction algorithms(PCA, Isomap, and LLE)in the task of spot weld quality diagnosis. Results show that the proposed method improves quality diagnosis accuracy in the abnormal condition test set from 83.18% to 93.64%. The root mean square error (RMSE) for nugget diameter prediction is reduced by nearly 0.33 mm compared the other four methods, effectively enhancing the generalization ability of the quality diagnosis model for out-of-distribution data.

Key words: resistance spot welding, online quality diagnosis, attention mechanism, domain adversarial network, out-of-distribution generalization

中图分类号:

TG441

宋强, 王闻杰, 贺晓斌, 夏裕俊, 李永兵. 点焊质量诊断模型的泛化性能研究及改进[J]. 机械工程学报, 2025, 61(18): 64-75.

SONG Qiang, WANG Wenjie, HE Xiaobin, XIA Yujun, LI Yongbing. Research and Improvement of the Generalization Performance of Spot Welding Quality Diagnosis Model[J]. Journal of Mechanical Engineering, 2025, 61(18): 64-75.

参考文献

[1] CAMERON S，PAUL C，MATTHEW D. A comparison of resistance spot weld quality assessment techniques[J]. Procedia Manufacturing，2019，29：305-312.
[2] 杨淑利，梁潇，邵立民，等. 空间柔性太阳电池阵发展现状及趋势[J]. 宇航学报，2023，44(12)：1810-1819. YANG Shuli，LIANG Xiao，SHAO Limin，et al. Development and trend of spacial flexible solar array[J]. Journal of Astronautics，2023，44(12)：1810-1819.
[3] REDDY P K，JAIDI J，GUDE S D. A numerical study on contact conditions，dynamic resistance，and nugget size of resistance spot weld joints of AISI 1008 steel sheets[J]. Numerical Heat Transfer，Part A. Application：An International Journal of Computation and Methodology，2023，84(1/6)：122-140.
[4] ZHOU K，YAO P. Review of application of the electrical structure in resistance spot welding[J]. IEEE Access，2017，5：25741-25749.
[5] DICKINSON D W，FRANKLIN J E，STANYA A. Characterization of spot welding behavior by dynamic electrical parameter monitoring[J]. Welding Journal，1980，59(6)：170-176.
[6] CHANG H S，KWON H C. In-process monitoring of micro resistance spot weld quality using accelerometer[J]. Korea Journal of Welding and Joining，2011，29(1)：115-122.
[7] XIA Y J，ZHOU L，SHEN Y，et al. Online measurement of weld penetration in robotic resistance spot welding using electrode displacement signals[J]. Measurement，2021，168：108397.
[8] ZHAO D W，WANG Y X，LIANG D J. Correlating variations in the dynamic power signature to nugget diameter in resistance spot welding using Kriging model[J]. Measurement，2019，135：6-12.
[9] CHEN G，SHENG B，LUO R，et al. A parallel strategy for predicting the quality of welded joints in automotive bodies based on machine learning[J]. Journal of Manufacturing Systems，2022，62：636-649.
[10] ZHAO D，BEZGANS Y，WANG Y，et al. Performances of dimension reduction techniques for welding quality prediction based on the dynamic resistance signal[J]. Journal of Manufacturing Processes，2020，58：335-343.
[11] KIM J，PARK Y，KU N. A study on the machine learning method for estimating resistance spot welding button diameter using power curve and steel type information[J]. Mechanical Science and Technology，2023，37：3711-3719.
[12] XIAO M，YANG B，WANG S，et al. A feature fusion enhanced multiscale CNN with attention mechanism for spot-welding surface appearance recognition[J]. Computers in Industry，2022，135：103583.
[13] ZHOU J，XI Z，WANG S，et al. A real spatial–temporal attention denoising network for nugget quality detection in resistance spot weld[J]. Journal of Intelligent Manufacturing，2024，35(6)：2743-2764.
[14] CHANG F，ZHOU G，DING K，et al. A CNN-LSTM and attention-mechanism-based resistance spot welding quality online detection method for automotive bodies[J]. Mathematics，2023，11(22)：4570.
[15] JIA P，SHENG B，ZHAO G. Machine learning-based algorithm for the classification and prediction of multi-type spot weld quality[J]. Journal of Advanced Mechanical Design Systems and Manufacturing，2023，17(6)：0074.
[16] LYU T L，XIA Y J，LI Y B，et al. Quality prediction and model explanation of resistance spot welding process under varying fit-up conditions[C]//International Manufacturing Science and Engineering Conference. Vol. 87240. American Society of Mechanical Engineers，2023.
[17] BOGAERTS L，DEJANS A，FAES G R，et al. A machine learning approach for efficient and robust resistance spot welding monitoring[J]. Weld World，2023，67：1923-1935.
[18] 刘振宇，张楠，裘辿，等. 基于小样本数据增广的产品服役性能预测方法[J]. 机械工程学报，2024，60(6)：11-20. LIU Zhenyu，ZHANG Nan，QIU Chan，et al. A method for predicting product service performance based on small sample data enrichment[J]. Journal of Mechanical Engineering，2024，60(6)：11-20.
[19] GANIN Y，USTINOVA E，AJAKAN H，et al. Domain-Adversarial Training of Neural Networks[J]. Journal of Machine Learning Research，2016，17(1)：2096-2030.
[20] 乐珂，李霁蒲，陈祝云，等. 基于无源自监督域适应的旋转机械跨域新故障诊断方法[J]. 机械工程学报，2025，61(3)：67-76. LE Ke，LI Jipu，CHEN Zhuyun，et al. A new fault diagnosis method for rotating machinery across domains based on unsourced supervised domain adaptation[J]. Journal of Mechanical Engineering，2025，61(3)：67-76.
[21] 邵海东，林健，闵志闪，等. 分布外样本干扰下基于改进半监督原型网络的齿轮箱跨域故障诊断[J]. 机械工程学报，2024，60(4)：212-221. SHAO Haidong，LIN Jian，MIN Zhishan，et al. Improved semi-supervised prototype network-based cross-domain fault diagnosis of gearboxes under distributed out-of-sample interference[J]. Journal of Mechanical Engineering，2024，60(4)：212-221.
[22] XIA Y J，SU Z W，LI Y B，et al. Online quantitative evaluation of expulsion in resistance spot welding[J]. Journal of Manufacturing Processes，2019，46(10)：34-43.
[23] XIA Y J，LI Z R，WANG W J，et al. Influence mechanism of initial gap disturbance on the resistance spot welding process[J]. Automotive Innovation，2024，7(2)：360-372.
[24] DONG G G，LIAO G S，LIU H W，et al. A review of the autoencoder and its variants：A comparative perspective from target recognition in synthetic-aperture radar images[J]. IEEE Geoscience and Remote Sensing Magazine，2018，6(3)：44-68.
[25] LI H G，TROCAN M，SAWAN M，et al. Serial decoders-based auto-encoders for image reconstruction[J]. Applied Sciences，2022，12：8256.
[26] CHUNG J，GULCEHRE C，CHO K，et al. Empirical evaluation of gated recurrent neural networks on sequence modeling[J]. Neural and Evolutionary Computing，2014，12(11)：1-9.