汽车工程 ›› 2024, Vol. 46 ›› Issue (2): 366-374.doi: 10.19562/j.chinasae.qcgc.2024.02.019

• • 上一篇    

铆接铝合金板铆钉失效缺陷检测方法研究

刘凉1(),张滢1,史晨阳1,赵新华1,孟宪明2,刘增昌3   

  1. 1.天津理工大学,天津市先进机电系统设计与智能控制重点实验室,天津 300384
    2.中国汽车技术研究中心有限公司,天津 300300
    3.中国汽车工业工程有限公司,天津 300113
  • 收稿日期:2023-07-25 修回日期:2023-08-23 出版日期:2024-02-25 发布日期:2024-02-23
  • 通讯作者: 刘凉 E-mail:liuliang_tjut@126.com
  • 基金资助:
    国家重点研发计划项目(2017YFB1303502)

Research on Detection Method of Failure Defects of Rivet on Riveted Aluminum Alloy Plates

Liang Liu1(),Ying Zhang1,Chenyang Shi1,Xinhua Zhao1,Xianming Meng2,Zengchang Liu3   

  1. 1.Tianjin University of Technology,Tianjin Key Laboratory for Advanced Mechatronic System Design and Intelligent Control,Tianjin  300384
    2.China Automotive Technology & Research Center Co. ,Ltd. ,Tianjin  300300
    3.Automotive Engineering Corporation,Tianjin  300113
  • Received:2023-07-25 Revised:2023-08-23 Online:2024-02-25 Published:2024-02-23
  • Contact: Liang Liu E-mail:liuliang_tjut@126.com

摘要:

针对车身用铝合金板内部铆钉缺陷特征提取难度大、缺陷类型与程度识别准确率低的问题,提出一种基于高斯卷积深度信念网络与双向长短期记忆网络相结合的铆钉失效缺陷诊断模型与检测方法。首先,面向5种铆钉断裂缺陷设计试件并搭建自动检测系统,通过规划和调整探头姿态有效地降低提离效应对检测信号的影响。其次,设计双网络融合诊断模型提取和学习多维度缺陷特征信息,解决检测曲线中由时序变化特性和空间分布状态表征的缺陷信息提取难题。实验结果表明,与传统卷积网络及单一深度信念网络相比,优化后算法诊断模型的平均准确率为99.85%,相比提升了14.54%,且具有良好的通用性和鲁棒性,可实现铆钉内部缺陷的在线诊断。

关键词: 铆钉内部缺陷, 检测系统, 模式识别, 特征融合

Abstract:

For the difficulties in feature extraction and low recognition rate in defect types and grades of rivet on aluminum alloy plates for car body, the diagnosis model and detection method for rivet failure defects are proposed based on the Gaussian convolutional deep belief network and long short-term memory network. Firstly, the specimens are designed for five types of fracture defects and an automatic detection system is constructed. The planned path and pose of the probe are set to lower lift-off effect on signals. Secondly, the dual network fusion diagnostic model is designed to extract and learn the multi-dimensional defect feature information, solving the problem of extracting defect information represented by temporal variation characteristics and spatial distribution state in detection curves. The experiments results show that the optimized model has an average recognition rate of 99.85%, with an increase of 14.54% compared with that of the traditional convolutional network and single deep belief network. The model has better compatibility and robustness, which can realize online diagnosis of internal defects of rivets.

Key words: defects in rivet, detection system, pattern recognition, feature fusion